Forgot your password?
typodupeerror
Slashdot.org News

Suggestions for a Startup Web Company 209

Posted by Cliff
from the taking-that-first-doosy-of-a-step dept.
mochaone asks "I've always admired the Slashdot crew for putting together a great site that has vastly contributed to the internet experience. I have an idea for a website that I think has great potential also. I would like to know how slashdot (or any other webcompanies) got started and what tips they might offer? Should I use webhosting services or provide content on my own computers? What's a typical server setup -- separate boxes for web servers, database, banners, etc? T1 line or T3? How often should I backup data if providing content on my own computers and should I store backups offsite? Any other tips are welcome. More interested in the high-level, architectural issues rather than the "Use Debian over Redhat" or "Use Python over Perl" issues. I think those have been covered in other Ask Slashdot features. "
This discussion has been archived. No new comments can be posted.

Suggestions for a Startup Web Company

Comments Filter:
  • One thing to consider is that it's really easy to provide the "simpler" network services (DNS, basic routing, that sort of thing) on separate, low-powered boxen (you know, all those old '486s you've got laying about.) That way, a single failure is less likely to take down a bunch of services - and, you've off-loaded some system load.

    Oh yeah, first post :-P

  • Right now I'm letting a web provider take care of keeping the servers up for SaveTheLaptop (http://savethelaptop.zzweb.com/). It's doing fine even though it's running MySQL, Apache and other applications on one server. The server is also situated on a DSL connection and is working fine.
    If your site isn't expecting to get much traffic then you don't need a high-end connection, but if it does get popular enough, it would be best to host it yourself. And yes, storing backups ffsite is recommended if your server will be located in a high-risk area (like California for example)

    --
    http://savethelaptop.zzweb.com/: Tips and help for laptop theft
  • If you know what you are doing, and have the money, run it with a T1 line though your own boxes. If you don't have a clue or don't trust your own experience, webhosting is the best way to go.

    It also depends on how big you really want it to be. Is it going to be huge with thousands of hits an hour? or reletively small with only a few hundrend hits a day?
  • On the subject, what is the best place to start when looking into advertising on a new site? Or even if the site has begun generating decent traffic, where does one turn to find the best advertisement revenue? The 3rd party agencies all seem to pay incredibly low dividends.
  • by JediLuke (57867) on Tuesday November 16, 1999 @02:36PM (#1527202) Homepage
    You have to lookat what everyone has and not do that. It is something people have already seen, thats my /. and mp3.com are so successful. You have to think of something origional and neat to do. I recommend a T1 with a nice server to use. A suggestion is to use a raid 1 so that you can pull one drive just in case, and you will always have a backup. Definatly seek some hosting and co-locations beacause you can charge and pay the fixed costs with that stuff. Then you can think about dial-up and dsl hosting as my company has. You aren't going to get very much from dial up so it is almost not worth doing...i might be biased towards dsl/co-location/hosting. brings in more money to cover your bottom line.

    Good server setup:
    Pentium iii 400-500
    256-384meg of ecc sdram
    adaptec aaa-131 (1 chan) raid card or the aaa-133 for 3 chan.
    raid 1 or 5 depending on your space needs
    and a t1, go with one t1 then multiple t1s then think about a t3. Unless you are getting hits like /. you prolly won't need a t3.

    that should be enough to get you started, and not have to upgrade the actual server for a while.
    JediLuke
  • by toofast (20646) on Tuesday November 16, 1999 @02:37PM (#1527204) Homepage
    I use different boxes for different type of content: ordinary Pentium 100's for the static pages and PII-350 + for dynamic content (perl, c++, asp). You could add a server just for images and graphics.

    This type of load balancing is real easy, but it's not fail-safe: if one server dies, all the site is affected. That's why I have a few old boxes with replicated content that are ready to rock in case a server dies. All I need to change is the IP.

    A T1 line is not very hard to saturate, and to some extent, only one server could probably manage the whole static/dynamic/graphic content.

    We have a T1 line and host our own content. This way you get much better control. We use one tape backup per server, and investigate our logfiles as to when the best time to backup is. That's typically 3am for us.
  • Until recently I thought that just keeping up to date with patches on linux would be enough! NO WAY !

    Whatever you do, setup a firewall. I am currently working on setting up OpenBSD on a machine to run as a firewall.... and if you are going to run a business, i would recommend that you keep funds aside for a firewall and work on security.
    On that end i would think that having a database on a different server would be nice. Preferably one that does NOT have a direct connection to the internet!!! (Hmm maybe it could be hooked up via the NEW USB standard :)

    Also, make the website that works well and dont PUT TOO MUCH into it. I think a lot of ppl. get annoyed when you can see 6 guzillion things and still can't navigate!

    :Lastly, try and get your website to be featured somehow onto /. and pray that your setup survives!

  • Getting your own lines is probably not a good idea. It's too expensive, you'll end up paying for bandwdith which at the beginning you won't need and it may not scale well if your site starts generating a lot of traffic. Which, few sites actually do. Generic webhosting isn't a good idea either, not flexible enough. Your best bet is colocation - use your own boxes connected to some ISP's bandwidth. Jump.Net is good in the central Texas area.
  • And I mean _get_it_now_. Don't wait until other pieces have fallen into place - by then your domain's probably long gone.

    Also, when you've settled with a fabulous, unreserved name, don't mention it anywhere until it's yours. I made that mistake once and the name was taken by a speculator in a matter of days... grrr...
  • That you need to be clearer about what it is you want to do. What services are you going to provide? What's the hook? How are you going to draw people in? What services/equipment you use completely depends on exactly what services you want to provide, how much you are charging for them, who your clients are, etc..
  • I think that the bandwidth, distribution of load, etc are really dependant on how big u plan this site to get and how fast. Also its going to depend on what kinda capital you have.

    I've been looking to start a database driven site for a bit and am putting a lot of thought into the architecture of the database... like, how i should organize tables to give me the best speed for queries.

    Another question that u might want to look at (and a question that im wondering about) is what kind of processor works best for what u want to do. I've heard risc based processors work badly for floating point... but what if yer doing a lot of integer operations? (i duno where this would come up...)
  • by sporty (27564) on Tuesday November 16, 1999 @02:42PM (#1527211) Homepage
    • Co-location vs dedicated own
    • Much bandwidth can incur more charges in a co-lo
    • Co-lo's have less responsibility for facilities, line maintenance, etc...
    • Backups

    • Daily if not half daily. Do full backups at least once a week and incrementals in-between. If you could do full's everyday, all the better. If you can also do redundancy, that would kick butt too, but redundancy (what?) redundancy (what?) is no replacement for backups. Take backups offsite. A must.

      As for services, have a different host-name for each service... so if you wanna move your database off of your web machine, www.myhost.com still hosts the web and db.myhost.com still hosts databases.

      Make sure to use the latest (or at least clostest) to the newest stable unix you can. Make sure you keep your software up to date security wise and feature wise when it is intelligeble. (If we are on HTTP 1.2, you should probably upgrade to keep up on standards.)

      Other tips, make sure you have a policy for the machine. Even if its for 3 people working on the machine, if rules aren't established early on, chaos may ensue when conflicts arise or the company grows. Most important, be openminded to learning. But you seem to have that. ;>

      ---

  • Why A P3? How exactly will SSE help a web server? Personally, I think RAID is overkill when you are just starting. And your server hardware depends a lot on the content you will be serving. If you are going to have a lot of static content, a fast HD is important. If you have more dynamic content (e.g. slashdot), then CPU power becomes more important. In the case of dynamic, you might want to consider dual CPUs.
  • I can't say because I'm not close to releasing to the public yet. Possibly in about 6 months or so. I can say that it's not really in the geek realm. At any rate, it's great in my mind!
  • hardware, yeah i would reccomend separating the db from the web servers regardless of weather you host it or not.

    if your comfortable, or you think you could get comfortable, with considering all thats involved with keeping a site up, ie backups, security, reliability etc. go for it. with remote hosting you lose control. it gets hard to try out new stuff, because your not the administrator. but you do have someone to yell at when it goes down...

    as far as bandwidth goes, it's far easier to go up to a t3 from a t1, as opposed to getting rid of a 15-45K/month phone bill for that fat t3.

  • I don't think his plan is to give everyone access to his idea so they can do it first. I think his plan is to find out the best way to impliment his idea, so he can do it himself.
  • Oops, hit the submit button by mistake. This version has a few corrections.

    The Zope [zope.org] web site lists serveral Zope Hosting Providers [zope.org]. Ask one of these ISPs to install the Squishdot product and you can have have a ready-made slashdot up and running in no time. If you need something more complex you can buy the premium service and install your own custom built products.

  • Obviously your server architecture should be driven by your expected load, but you should also design your app for scalability(meaning you can add boxen easily).

    For the companies I typically work for its split into a static image server, a couple apache/modperl servers, a big database box, and possibly middleware box(en) for caching data. Add a sendmail box if you expect heavy email traffic. Possibly an admin box on a VNC.

    But again this design is for a biggish site. Sometimes just two boxen (apache on one and database on the other) will do just fine. I would try to avoid putting them on the same box just for security.

    Colocation seems like the way to go, you get control of your own machines, but won't have to worry about power and pipes. This may be colored by the bad experience I've had with one of those full service type shops.

    Backup: What else am I gonna say? Backup to tape daily, with off and onsite redundancy. You may just have to back up your database daily(if all your content lives there), and just be ready to rebuild your web server if it goes south.

  • by deicide (195) on Tuesday November 16, 1999 @02:45PM (#1527219)
    for a 2-3 hundred/month you can get a dedicated hosted server from someone like dn.net (where btw slashdot is hosted). You will have access to massive bandwidth and full control over the box. You could also find a (semi)local company that provides similar services in which case you would have physical access to the server.

    Even load balancing between 5 boxes hosted at dn.net will probably cheaper than running a dedicated T1 line to your office.. And your servers will be on multiple-DS3s..
  • Do you have any more information on colocation? I'm not sure how I could use an ISP's bandwith while using my own boxes without having a line run.

    Thanks.
  • Oops, hit the submit button by mistake. This version has a few corrections.


    The Zope [zope.org] web site
    lists serveral Zope Hosting Providers [zope.org]. Ask one of these ISPs to install the Squishdot product and you can have
    have a ready-made slashdot up and running in

  • think about it - if he did have a plan - he wouldnt be airing it first - otherwise it may be corporatized. Grade school logic says that if he doesnt give any specifics, he doesnt necessarily have no plan
  • I own a webhosting co, and we've had a few clients that started with us but then after geting much bigger than they had planed move to a seperate option, be it their own setup or something like colocation.

    I think really there are three options for people:
    1) simple webhosting if you aren't planing on having too much traffic (this is sharing your server with alot of other sites, after all)
    2) CoLocation If you're planing on alot of traffic but don't have alot of money/time to deal with firewall/server/bandwidth/etc/etc/etc issues
    3) Roll your own If you're expecting alot of traffic (slashdot-esque) and you have the resources (cash, knowledge, possibly staff)


    SE [sevenelements.com]
  • Colocation (aka co-lo) means exactly what it sounds like: You take your machine and set it up in their server room. They give you an ethernet plug that goes to their Internet backbone, thus putting your box on the net through their connection.

    This may be adventageous for you, in that you don't have to maintain a constant temperature and its someone else's neck if the connection dies. Just make sure they allow you physical access to the box when necessary (think upgrades), and have some kind of ability to either reboot for you or allow you to do your own remote reboots (think system crashes).

    --------------------

  • RAID may be overkill for a small website, but outages are not acceptable if you are small or big. My opinion is the best method is at least two webservers with synchronized content with a round-robin DNS entry. This way even if one physically explodes, the other can manage fine.

    On the other hand, RAID never hurts. RAID is our friend :)

    I will totally agree that a PIII is not needed. A few P-233's should do the trick for a while.

    One thing that most (good) ISPs will do is be the primary nameserver for your domain with your nameserver as the second, but with updating going from the second to the primary. This keeps most nslookups off your subnet, reducing traffic. Did that make sense? Damn, I hope so :P
  • I hate to say it, and I don't want anyone to think I'm ripping off slashdot, but it's going to be forum-type environment (prefer not to say what the content is....not porn), therefore it's going to be database intensive and web server intensive (lotsa dynamic web pages). I'm not charging for using the services and my clients will be anyone who wants to participate.

    I just have an idea that I want to unleash and I'm not really concerned about making an internet million. I have some money to play around with so I'm trying to get feedback on what things to think about before diving into the deep end. Most of my friends are sports nuts and don't have a clue about geeky stuff so slashdot is my best bet on getting some info.

    Thanks.
  • The companyI work for has machines colocated with quite a few datacenters out there.

    The way it works is you rent space in their datacenter. Either a cage, or part of a rack depending on how much equipment you have. You supply the machine, and pay for bandwith, and they will provide you with hands and eyes support for your machines. You can usually get into the datacenter as well, assuming you colocate the machine someplace close to your house.

    This saves the cost of running a line somewhere, and it gives you more of an options on the bandwidth you want. You can start off with a smaller amount of bandwidth, and upgrade without having to change your equipment (assuming you're not trying to get 11Mbps on a 10Mbps nic card, or 110 on a 100Mbps card)

    It's a rather nice setup, but you do need to watch out for who you colocate with. Some of them can be real pains to deal with. I just spent an entire weekend trying to get a machine in a New York datacenter rebooted because the oncall tech for the ISP (which shall remain nameless) wouldn't awsner pages.
  • ALWAYS keep off-site backups.

    The risks of physical location are not important -- well, so long as you don't co-lo on a phone pole in a dark alley on the bad side of town. Hard drives DO fail; power spikes DO occur; rats, mice, slugs, roaches, etc. etc. ...

    And then there's the "you can have the box back when it's been deguased" situations. [rare, but you're only paranoid until you can prove everyone's out to get you.]
  • You just pay a monthly fee and slap your box into your ISP's lil network in their building. They'll give you an IP and a DNS to use for when you register your new site and all.


    spiffy
  • I'm a part-time webmaster, so I use virtual hosting with several different ISPs. My favorite is Pair (www.pair.com [pair.com]). They don't do any dial-up services (read: bandwidth hole), run BSD servers (which I prefer to Linux for web hosting), and I get to let them take care of the details. That way I can take off for the weekend and not worry about my server going down while I'm gone. Yes, I sacrifice some flexibility and power, but my time is valuable. Matt
  • by Anonymous Coward on Tuesday November 16, 1999 @03:02PM (#1527236)
    If you're really serious about getting into web stuff, study http://arsdigita.com/ [arsdigita.com] and http://photo.net/ [photo.net], as there's stuff like an entire book [photo.net] online on how to set up a serious e-commerce site. And the author's an entertaining writer, too.
  • Crap! not rug burn! There goes my startup. Forget about that IPO.

    :p

  • Not a chance, dude... *I* invented the hydrogen atom (and licensed it to God), and I've got the documentation to prove it. If you disagree, my lawyers and I will see you in court. Unless, of course, you'd care to come to some agreement... (cackles evilly)

    Slarty
  • Personally, I think RAID is overkill when you are just starting

    I disagree - RAID doesn't just give you data protection, it also buys you availablity. A hard drive failure in a non-RAID system will take your server down temporarily. Even if you didn't lose any data, it still sucks. If a drive fails at 3am, what would you rather do: get paged ("Hey! the server is down. Go fix it."), drive in, swap hard drives, restore server, try to get back to sleep - or - drop in a fresh hot-swapable drive when you get in the next morning? I think you can guess my answer.

    Disclaimer: RAID is not a substitute for load balanceing on multiple servers and/or proper network design

  • by mr (88570) on Tuesday November 16, 1999 @03:04PM (#1527241)
    You need to answer basic business questions 1st.

    What is the market you are serving?
    Why are customers going to come to you, and keep coming back to you over others?
    Does anyone in your group of people understand basic accounting?
    Do you know how to run the accounting package you have chosen?
    How are you going to link this accoutning package to your business?

    And the REALLY big one:
    Do you have the money to do this?

    Sweat, blood and credit card advances only go so far.

    Go to the public library, any of the small business web sites, and even (gasp) the IRS and do some reading. They will tell you ALL kinds of questions you should be asking. Like Insurance, type of business org., etc.

    After you have done the above mentioned research *THEN* start wondering about DS1 or DS3, colocation, etc.

  • Power consumption and heat generation for two. Plus, soon you won't be able to find any PII's around.

    I would agree RAID is overkill for a startup. What little speed and protection you want from RAID can be done just as well and sufficiently fast in software (just about any OS can do software RAID.)

    My advice is to keep backups more than investing time or effort in RAID. Of course, if you need 100% availablity, then you are beyond the average /. commenter :-)
  • by Signal 11 (7608) on Tuesday November 16, 1999 @03:09PM (#1527243)
    How to "work the web" for dummies:

    • Patent everything. If you can use a mouse with it - patent it.
    • If your product sucks, blame Microsoft. Even if it doesn't, Microsoft is the reason nobody knows about it.
    • Use atleast seven of the following buzzwords in any promotional material. Super PHBs comb for this stuff via keywords, and you DO want your website to show up, right? Paradigm, proactive, "think outside the box", revolutionary, third-wave, interactive, multimedia, 3D, any word with i, e, or x in it. For example: iBrain, e-data, or Xtreme.
    • Your HTML code should suck - amazon.com, microsoft.com, networksolutions.com, yahoo.com - all their HTML sucks. Yours should too.
    • Get vulture capital, and then immediately go to IPO (see this [linuxone.com] example) without creating a product. Claim your business is riding the 'bleeding edge' of technology, and products are obsolete - you sell ideas, not products!
    • Run linux. Hey, with Microsoft on the out-and-out, it helps to run an OS loved by millions. Try to get it linked to slashdot too. Get an interview if they won't post it and convince them you have a new "open source" methodology to designing websites!
    • Marketing, marketing, marketing. You can't go wrong with huge banners proclaiming you're THE hip business to do business with on the 'net. Just don't claim you invented the internet unless you're running for president. Everything else is 'OK'.
    • Whatever you lack in content - make up for it in huge flashing banner ads and broken HTML that only renders correctly once in a blue moon. Web surfers love to see people using technology so new they can't even view it!


    --
  • by Anonymous Coward
    Advertising in general is not effective on the web. Click through rates are declining, and there is research that shows that users' eyes don't even track to the banner ads. Advertising is a losing proposition in the long run. See www.useit.com [useit.com]
  • Actually, if you do round-robin DNS rather than intelligent redirection, you aren't okay if one machine goes down - you're dropping a full half of your traffic, right?

    Seems to me that if you're going to go to the trouble of redundant servers you might as well add some smarts to make the best use of them.
  • by Anonymous Coward
    It's more than annoying at the massive chasm in pricing that exists between commercial and non-commercial pricing for any sort of high speed internet access, even when comparing identical speeds. There just isn't an affordable way to start small and grow.
  • I've gone through this myself. I wanted to start a hosting business. I got an ISDN connection into my basement and set up a couple of linux boxes. The ISDN (from UUNET) cost $600/month for the feed plus $120/month for the line. I never broke even.

    Eventually I moved to co-located with 9netave.com and now I'm making money. I positioned myself as a low cost hosting company and automated everything to make it so that it wouldn't drain more time than it was worth.

    The moral of the story is to start with something you can afford, and expand as required. Since you're just building one site I'd start with a hosting company and build and test it there. It would suck to pay all the money to set up a co-lo and then have the thing fizzle. (Short plug: I support php3 and mysql plus 100MB storage for $20/month [zymsys.com] - Sorry - I had to!)

    If all looks good, move it to a co-locate and try to go big-time. If that takes off then maybe you can afford your own bandwidth, but it's tough to compete with co-location.

    As my business grows it may become less expensive to run my own T1, but then I'm loosing the speed of all those OC-3's, so it'd probably never happen.

    Hope this helps.


  • Any natural disaster big enough, like the current series of earthqueakes in Turkiye, can make your whole physical location just disappear in a cloud of dust.

    With either good friends or severe costs, you'll be able to replicate the whole service on a server on another continent, ready to jump into action in the worst case. The DNS will take a while to update :I

  • by Eg0r (704)
    they're the people who host Thomas Pabst's site tomshardware.com [tomshardware.com]

    They must have plenty of bandwith if they can handle it! I'm sure Tom explained how he got to choose pair, but I can't seem to find the link (and Thomas the searchengive is dead...)

    ---

  • You haven't given us much to work with :) To answer most of those questions thoroughly we need to know more about your idea. I can understand you not wanting to announce it publically, but then you must understand our inability to respond accordingly. I'll try for some generic answers.

    How to get started? A slashdot article can always help :) Really depends on your audience. If it's just a site for you and your friends, tell your friends. If it's a humor site, try advertising on well known humor sites. If it's a geek site, get slashdotted and you're all set :)

    Typical server setup? There isn't a generic typical one. It depends GREATLY on what kind of traffic you expect. Slashdot and Andover probably have 30 billion dedicated servers, other, smaller sites, just need geocities or the like.

    T1 or T3? What kind of question is that? Of course you want the fastest connection possible. Maybe your real question is whether it's really worth the extra money to get a T3 instead of a T1. Well, a T1 and a T3 doesnt mean much - you could get a T3 with insane amounts of traffic and be slower than a T1. And of course, there are other choices besides T1 and T3 :)

    Backing up - definately key. I backup every 1 to 2 weeks. Note, make sure you dont overwrite your previous backups with new backups - I've done that :)

    Webhosting service vs own computer? Uh... depends how much cash you have and how good your computer and connection are. If you have a really fast internet connection and your ISP doesn't mind, then use your own computer. But if you dont know much about server admining then it's probably easier just to go with a webhosting service. At least to get started. Move from there.

    I think that covers it. Have fun - starting new sites can be a blast :)

    "Then I'll tell the truth. We're allowed to do that in emergencies."

  • I can't agree enough about the necessity of backups. Redundancy is great (crucial, for when you want to colocate or start doing real load balancing) but it doesn't address the situation of a terrorist bomb, the FBI, an earthquake, a tornado, a hurricane, etc. of wiping out your whole site. With offsite backups, you can reinstall your data on some new machines and be up and running again.

  • Hmm, let's see. To have a web site, I think you need a computer. You should get a computer, or find somebody to loan you theirs.

    Seriously, the only clue you've given us is that your web site "has great potential" (what doesn't?) and that it might have some use for banners and a database. That's a lot like saying "I have an idea for a computer program that has great potential; what language and OS should I use?"

  • Where I live, MediaOne doesn't provide static IPs and you aren't supposed to run a server off of the line (and their service sucks). At work we host a site off of a 1.5 Mbps DSL line pretty well. But....we also pay several hundred a month compared to 49.95 for the cable modem.
    Cable modem vs. DSL technically shouldn't matter. It all depends on Speed/Price, availability/cost and availability of Static IP addresses.
  • haha, so true. a good porn site will beat any ol' e-something or other site.
  • by nacho (1714) on Tuesday November 16, 1999 @03:28PM (#1527257) Homepage
    I've seen a lot of posts here regarding T1s or T3s. I've done a bunch of research into this, and it breaks down into this:

    You pay for your local loop (from your local telco) from your location to your service provider (UUNet or Qwest etc.). That typically costs around $350-$450 for a T1, and a little more expensive for a T3. Then you've got your net connection charges. You can get a fractional T1 or T3, and have less bandwidth to use, but you're not paying as much. When you need more bandwidth, give 'em a call, and they can up it for you.

    That doesn't work so much with T1s, 'cause it really doesn't make sense price-wise since theres a lot of money involved in the connection itself, regardless of the bandwidth.

    I've found Qwest to be the cheapest solution, and I hear the quality is really good (guarenteed 100% uptime, and 0% packet loss...within their network). What you could do, is purchase a T3, but only pay for 3megs of bandwidth (its usually charged in increments of 3 megs). Need more bandwidth? Add another 3 megs.

    There are other solutions (especially for T1s) called Burstable T1s. This is where you only pay for an average bandwidth...for instance, they monitor your bandwidth usage every 5 minutes (in the case of UUNet), and average out your usage. They then take off the top 5% of your bandwidth, and then charge you for what you've used. The 5% off the top is so that, say for instance, you get /.'d once, but the rest of the month you're back to normal...you don't get charged for the /. (if it fits in that 5% usage).

    Anyway...My suggestion is, if you're expecting a lot of growth, get a router that can support up to 45 megs/sec (full T3), get a T3, but only pay for 3 megs to start. For me, I figured the monthly bandwidth charges to be around $1300 for a T1's bandwidth, and $350 for the local loop. I can't remember what the charges were.

    Lemme say again, that UUNet is EXPENSIVE! And, from an ISP standpoint (which is where my day job is), they're down a lot. I've never dealt with Quest, but they're pretty responsive to their customer service calls, and they boast a really sweet network.
  • by Anonymous Coward
    Instead of replicating the data across continents, just replicate the data on opposite sides of a city so large that if it were destroyed, nobody's going to care about your tiny website.
  • Not true (well, not entirely):

    ISDN
    Depending on where you are, ISDN can be damned cheap or damned expensive. You get a limited choice of bandwidth (56k up to 128k)

    IDSL
    Just about as expensive as ISDN but 16k faster *grin*

    SDSL
    WAY cheaper than a T1. DSL can run higher than T1 speeds depending on technology and distance to the CO.

    Frame Relay - Fractional T1
    Cheap compared to a dedicated T1 and you can change the bandwidth up or down with relative ease.

    Frame Relay - Full T1 w/CIR (burstable)
    Also cheaper than a dedicated T1 and gives you most of the bandwidth of a T1 most of the time depending on the area, Telco, and ISP

    Colocation is an odd bit of voodoo. Finding the right match can be difficult for a startup. There are usually throughput limits on co-lo setups. And, colo can end up just as expensive as a full T1 pulled into your kitchen. Of course, there's also co-lo services out there cheaper than most dialup accounts :-) (I hate looking for colo space.)

    If I didn't know what my needs would be initially or what my growth would look like, I'd look at SDSL; it's not that expensive and provides a reasonable bandwidth (you may qualify for full T1 but only buy 384k and change it as you need to.)

    These things vary from place to place, so definately shop around.

  • As an overpaid consultant, setting up databases for websites in silicon valley, I'd like to say this:

    Don't hire anybody just yet. Pay as little money as possible, and get as simple and cheap hardware as possible. I'ts easy to talk about what the ultimate setup is, but you are at point A, and that is point B. You have to think in terms of your first steps, while visualizing the *path* to the larger setup.

    0.) get a domain name.

    1.) Check out www.cobalt.com. Look at the Qube or raQ server appliances. The skinny: simple setup, 90% of the features you will need ultimately, and 120% of the features you are capable of using today. That way, you get started on content, not on DNS, Apache, TCP configuration etc. etc. I can setup all these things, but if I was starting up my own website, that's exactly where I'd start, with an appliance like this.

    2.) get your site going in your bedroom, completely offline.

    3.) find a ISP that rents out cobalt servers. I think there are quite a few, at least in the bay area. Install your server there.

    4.) revel in the massive response to your site.

    5.) hookup with someone who can setup a faster linux box for you , and improve you site to eclipse slashdot.

    6.) hire another guy to do grunt work like backups etc. thay you've been doing till now.

    6.) roll the multi-server, load balanced version to a co-location facility, using the stuff others have talked about in this thread. 7.) work on your keynote address, focusing on the story of how you got started with a post to slashdot.
  • For now, I'm running a startup outta my home - but it's only in the devlopment phase. Some things I've budgeted for this one when I get some $$$ (in 3 weeks!):

    E-com server: P3-450(X2), 256MB, RAID, >= RedHat6, Enhydra [enhydra.org], IBM JDK 118, Apache

    Db Server:Postgres (free+transactions), same hardware (ditto for Auction Db server)

    Auction server: same hardware, Apache, EveryAuction [everysoft.com]

    All running on a 100mbps switch. When the need arises, we'll move the hardware off to Qwest or somethin. All the above can be purchased for under $20k! Our whole goal for the business - software wise - is free and open source.

    We've come to the obvious conclusion that there's no need to run anything M$.
  • My personal webserver is a SPARCserver 1000 w/8 CPUs. It pretty much rocks to tell you the truth. I only paid $250 for the thing and it has 512MB of RAM in it which was in the thing when I got it. It runs Solaris 2.5.1 and also runs a few other services (ftpd, sshd, telnetd, etc). The "line" is a 4Mbps "wireless LAN" unit from TTI. I couldn't reach my employer's site which is around 13 miles away (there is a DS-3 there) so I built a an amplifier that pumps out approx. 8W on 915MHz into a pair of 33 element antennas. I am going to put some of this stuff up (amplifier schematics [ which should also work on aironet, zoomair, wavelan, you name it at 900MHz, etc) for some content which is about all that I'm currently lacking. :-) Don't think ya gotta use intel h/w though, the high end older sun stuff can be had extremely cheap and it performs pretty well for the money.
  • Sorry to go off topic, but this is a great opportunity to get some opinions from people who know how sites should be designed. Speaking of Ideas on how to build a great web site I am running a consulting service using a local provider to host my web site, but im not doing anything special. I want to add some more active content to it, can any of you guys suggest anything ? www.cubedd.com
  • NTL, the main cable company in the UK will fit an E1 line (2Mbps), and let you start at 128K and increase the bandwidth as required. They only need 10 days notice, so as long as you keep an eye on the amount of traffic you're getting, you should be able to plan ahead.

    Co-location is fine, but the cost (certainly in the UK) adds up if you want to have several machines hosted - eg. several load-balanced static servers, cgi servers, an image server, MySQL server, etc, which you'll need if your website generates a lot of traffic.

    Usually you only get 1 or 2 Gigabits per month for each machine and have to pay per MB after that.

    So, if it's going to have a low bandwidth requirement and won't have thousands of hits per hour, then just co-locate one machine with an ISP, otherwise go for a full set-up like Slashdot (there's info on their set-up in the Slashdot FAQ).
  • Exactly my point, it doesn't matter what kind of fortress your box is in, when it falls into a river of lava, it won't matter much how many generators, UPSen, or redundant T3's the site has if you don't have a backup that's not also in that river.

    This is my friend Murphy. I'll let you two get aquanted.
  • First things first, your servers should probably be as follows: Static Content - P-120 /w 128MB RAM Dynamic Content - PII-350 /w 256MB RAM

    Depending on the amount of space you need either use RAID1 or 5. That way, if one drive dies, you don't have to resort back to tape yet.

    You will also want a decent tape backup system. I prefer the Exabyte Mammoth drives as they are quick and hold 20GB native.

    As for you backbone connection, my experience at my current job (network admin of a small ISP) shows that you can actually run quite a bit of content on a single full T1. If you are expecting around 10,000 hits, a T1 should be fine. You will want to get some sort of firewall and you would be smart to use multi-homing T1s. That way, when one provider dies, you have a backup.

    As for co-locating and such, being that I work at an ISP, I have never done that myself and we currently do not have anyone co-locate with us. I can see benefits to co-locating, but physical access to the machine is sometimes necessary and you will want to check and see if you can get to the machine any time you want/need.

    Other than that, you shouldn't have any problems. Another nice feature to look for when getting a backbone provider is to see if they offer some sort of emergency notification system for outages.

    "Those of you who think you know everything are annoying those of us who do."
  • by Anonymous Coward
    Don't start a business. Start a normal, hobbist-style web site. If it catches on, expand. Don't load down any capital in it. As a rule of thumb, most of the good sites were hobbist-maintained and grew. Most of the sites that suck huge amounts of cash started as businesses and floundered.
  • by sporty (27564) on Tuesday November 16, 1999 @03:53PM (#1527268) Homepage
    Good server setup if you wanna put all services on one machine and get 1million hits a day serving static and dynamic content.

    One should learn to use vmstat, iostat, etc.. to figure out how things are limited. If its bandwidth, splitting off the service to another machine is advisable. If it is still a problem, some sort of load balancing would be nice.

    But I cannot stress enough hosting your mail on mail.service.com, web on www.service.com. Even if they reside on the same machine, it makes the transition unbelievably smooth.

    And remember to beta test what you are doing. Its unprofessional to work on the live servers too.

    ---

  • Your parents lied to you. They said money isn't everything. Money is everything. To be able to generate a successful website, you need money. No longer are the days when sites become popular by word of mouth or thru Yahoo's Cool Site. You need to either be /.ed or pay for advertising. Look up the Ask Slashdots on venture capital. While your idea may be great and your site may look awesome, it won't be worth it if nobody shows up. Just my two greedy cents.
  • The Linux Virtual Server project, at http://www.LinuxVirtualServer.org [linuxvirtualserver.org] would be a good place to start.

    You can get some really high performance out of it, you don't need anything better than for example a Pentium 100! and all the information you need on setting up a Virtual Server is on the site.

    Cheers,
    J

  • Don't forget about the
    UPS backed up by the
    UPS backed up by the
    UPS.

  • if you are acutally worrying about a terrorist bomb, the FBI, an earthquake, a tornado, a
    hurricane, etc. of wiping out your whole site
    you should really just beam your whole website(minus /etc/passwds?) into space, so therefore if the entire earth blows up and you survive you can continue your website. of coarse then you'll have to transulate your site to klingon, but that is another matter.

    matisse:~$ cat .sig
  • Guaranteed 100 % uptime eh? So what do they do when someone comes along and digs through the telco line? Ooops. The only way to have guaranteed 100% uptime is to have at least 2 connections to the internet, and big beeefy cisco routers to hold all the routing tables. But seriously, its much easier to use a 3rd party hosting service if you're that paranoid about uptime.
  • Not to be too picky, but the answers you want are what network consultants get paid $100-$200 an hour to answer. There's not ``one answer for everyone.'' I'd recommend visiting your library or bookstore and grabbing a couple O'Reilly books and other stuff from the same section and start reading. Call a couple ISPs near you (or not near you for that matter) and explain what you want to do, how much traffic you expect, and the like, and you'll notice a lot of data starting to approach an asymptote. And therein lies your answer.

    (Or, just gimme a call and give me two hours and a hundred bucks. I'll set you up real good. :P )

    -Chris
  • Hmm...

    A lot of comments about hardware and hosting here and almost none about design misstakes.

    I'll tell you a story: for more than a year ago I started collecting links to different Linux programming resources. Is started out with some static pages and got my self a domain (linuxprogramming.com [linuxprogramming.com]). After a while I needed a database to keep track of all the resources so started learning MySQL, because that was what my hoster provided. I didn't know any SQL nor any database theory. After many hours of work, I manage to get everything to work the way I wanted it to.

    But then I got more and more ideas. I came across a big problem: because I didn't know what I was doing when I created my base, I made alot of misstakes (especially in the design of the database). Te result was that I could not implement all the new stuff. This summer I decided to redo all of it. That is more than 4 months ago, and I'm still not done! (School and work takes a lot of time)

    Anyways, what I'm trying to say is that if you have such a great idea, create a base that is easy to scale. In my opinon, it is easier (and a lot more fun) to upgrade hardware och bandwith than correcting stupid design misstakes you did for 6 months ago...

    Good luck with that idea!!

  • Hrm, maybe I'm totally misinformed. I guess I need to actually try disconnecting a server :) I was under the impression that it would look it up again, as I've seen ftp clients look up one, fail, and grab the next. I've seen command line ftp do this for "ftp.us.kernel.org". Perhaps it's application specific. Anyone know how web browsers handle this?
  • Sweat, blood and credit card advances only go so far.
    the founders of cisco founded their company with their credit cards...but they also lost the company and missed out on the bumper share price rise :(
  • Also things I think are important, since I know the tech stuff:

    1. Required Forms
    2. The form of business (Partnership, LL Partnership, Proprietership, LLC, Corporation, Corporation [type]...)
    3. Where to get reliable patent/trademark searches, and yes, lawyers are the best
    4. Zoning
    5. City/County forms
  • Some types of content (static) are sometimes best not housed on-site.

    I know a lot of sites that simply house their graphics and stuff on seperate machines, not within their SOI (Sphere of Influence), while keeping the important bits, like HTML and dynamic content, within house.

    Static graphics are a key example. Why not move grahics and the like to key web server providers, and just reference them in the HTML instead of your own machine? You can even use multiple different servers (ie: different web hosting companies) to provide faster delivery again.

    If you want to get really swish with load balancing, you can keep a list of sites where your pages are stored, and use a simple "fetch check" script to see if they're online and providing content. You can then hand out the "working" list through a round-robin-iser (that rotates the usage on the list) and have it automatically integrated into your dynamic or static pages when they are handed out.

    Some people might say "what about round-robin DNS's and stuff?" Good idea, but think about one thing. What happens when 1/2 your machines go down ? You update the DNS right? How long will that take to filter out to the other DNS servers that have cached info? How long will your system be unusable from some part of the net? Having a system that polls and keeps an updated list is a very useful idea.

    (Would be really nice if you could have the client figure out where to get content from based on which systems give the fastest response time, but then you get all sorts of other issues to deal with.)

    Viola, you have a very fast, reliable, and effective way of providing static content from your site without using much of your bandwidth.

    BTW: Ever noticed images.slashdot.org isn't the same box as slashdot.org? *grin* Offloading of key content is a great idea isn't it?

  • I would advise anyone running anything significant to be wary of pair. The site I work on ( www.kraproom.com ) was hosted there for a bit, until we found out their interesting anti-spam policy. If you get one complaint, even without headers, they pull down your account for a week. In our case, we were even able to prove the malicous intent of the person who sent the complaint, but it fell on deaf ears.
  • by Hobbex (41473) on Tuesday November 16, 1999 @04:34PM (#1527284)

    And I order you and Slashdot to immediately seize and desist all usage of our internationally copyrighted phrase "... for dummies". We have court decisions backing IDG's exclusive right to address the dummies of the world.

    Though, given how in line your advice is with other IDG publishings, we may consider giving you a book contract for this one...

    -
    We cannot reason ourselves out of our basic irrationality. All we can do is learn the art of being irrational in a reasonable way.
  • Whatever you do... sue Microsoft over that damn paper clip!!! Argghhhhh!!!!!!!!!
  • by AIXadmin (10544) on Tuesday November 16, 1999 @04:37PM (#1527287) Homepage
    Before you start a business you need to do a lot of planning. Places like startups.com can help you with that. You need to devide whether you are going to be a Corp, Inc., or LLC. Do I want to give up equity in exchange for deniro. Do I have the money to start this? Do I have the money to support myself while my business is loosing money? Am I a manager or an engineer? Do I want to bring in experienced managment to help me run it.

    Rob was one of the lucky ones. He just had a good concept and worked hard. Most people aren't that lucky!8 out of 10 startups fail, and not in a blaze of glory. (Anyone remember Pen Computing.)

    =====
    There are alot of good resources out their:

    My favorite three:
    SBA Online: Small Business Administration
    - http://www.sba.gov
    My Favorite Venture Capitalist: Garage.com http://www.garage.com
    They have a ton of resources for those thinking about getting started.

    SCORE: Service Corps of Retired Executives
    You can talk to people who have been their done that. Over email or in person. There free too.
    Cheers,
    WFE
    ===========
  • I run webhosting services for other businesses, and my setup is thus:

    6 Ultra Sparcs 3 main servers and 3 replicated. The raid is shared for each pair, with an automatic failover running on a serial line. Each Raid is Raid 5 with a hot swap for 22 GB on each raid. I have a graphics domain, a db domain, and a main domain. All scripts run on the main server, the DB server runs as little as possible and all maint tasks are run through the Graphics server. Each Sparc has 256MB of RAM.

    All of these are running on an FDDI ring to the router.

    Firewall: Linux server on a crappy P100. Also serves as the Samba server and PDC for NT. Also rigged for auto failover to another P75 that is replicated on a daliy basis. (Cron)

    3 Win NT workstations for design and input. Nobody, and I mean Nobody! but me ever sees the backend of my site.

    Backups: Full backups every night to a DLT.

    I also have a Cron Job that knows which processes should be running, and if it can determine it, kills processes it does not expect. If it cannot, it send mail to notify me.

    Read all the Hacker FAQ's and find out where the obvious hacks are. The lamers that read these to get started are generally about 90% of the attacks you will see as an admin.

    3 T1 lines. I know that a T3 is a better way to go, but with the pricing here, it's about the same price to get 3 T1's, and if 1 line goes down, my site is still operable. I experienced an outage due to the Telephone company about 2 years ago that had my sites off line for 48 hours.The money I had to reimburse my customers more than justifies the decision.

    Startup: Well, luckily I got started before everybody thought they could make a million off the internet. It is all a question of how much money you have. What can you afford?? My initial setup was just a single Sparc 20, and I grew it from there. (And those were brand new then!) Draw up a business plan. Expenses out + 10% + capital investment yearly + time cost. Now look at this number and ask yourself if you can make more than this on your site. If not, scale back and redo the calculation. You need high margin to make it in this business.

    Selling space to local concerns is a nice way to defer some of the initial setup costs, and you won't be using all of that disk space to begin with. (I hope! =|:>) If your admin skills are up to it, I do not reccommend renting space, due to possible restrictions from the provider. It's thold computer joke, "How much for a really good computer?" "How much do you have?"

    Without knowing more about what you are trying to achieve, this is the best I can do.

    ~Jason Maggard
  • Really, determine your needs. Break the problem down into smaller pieces. Think top down design. Remember that certain technologies work better for certain tasks.

    Consider using a multi-tier infrastructure if you are interested in a quick dynamic site. Consider perl, PHP, and Zope, which are all great technologies. If you consider using Perl and Zope, you might want to consider using XML-RPC to get them to work together. This is what I am working with. Try, though, to design with the future in mind - if you design modularly, make sure that your site will grow with you and that you can replace certain components in your architecture as they start to not meet you needs. Sean

  • If you don't even know what kind of network you're going to need, T1/T3/OC*, colocation, etc., its highly doubtful you're ready to jump into this.

    The advice here is basically worth what you've paid for it.

    I'm serious - you need to do way more homework than what you've related.

    The "easy pickings" in the .com market have been taken - its going to be hard slugging from here on in, so don't think you can just set up a redhat box and start charging for banner ads.

  • In matter of hosting a web site, I don't think you should focus on the server. The content is what people will really look at. I have no idea what kind of machines /. is running on (but I know who makes them), and I don't care about load balancing, redandoncy or if their routers use BGP as long as I can read /. .

    Saying that, I think that whatever you use, get ready for scaling. I was surprised that nobody talked about NIS (Yellow Pages). It is a must to have on a network (Unix ... of course). Think about having a file server, independant from your web servers.

    There are a lot of techniques you can use to get ready for scaling, but in my opinion those two are the most important.

    Good day!
  • by Anonymous Coward
    Well I work for a hosting company doing the job of architecting some sites, mostly startups with no idea what they are doing on the hosting side. here is what we usually do for our customers and what I usually recommend. Every site is different so if you want more info just send me some e-mail, my address is at the end.

    We usually begin start-ups with a basic configuration of 2 Compaq 1850R's, any rackmount small 3U or smaller server is good to start with, make sure its rackmountable so it can be co-lo'd at most places. We choose these machines because of their small footprint and ease of maintenance. They can also pack dual processors, 1G ram, and enough for 6 18G drives. With hardware the smaller the better, its saves money on rack space.

    The front end machine is run as a web server,and an app server if you have an application. The second machine is used as a backend database and it is on a private backend private LAN. This makes it easy for the web server to access but harder for hackers to get to it. We have a remote access solution to administer all of these machines from a secure backend network connection. We scale this configuration by first adding a webserver if traffic warrants. We load balance the webserver usually with hardware based load balancing, Cisco LocalDirector, Arrowpoint, F5 Labs etc etc. This works quite well and has very good performance. FOr the database Server we usually will add another for failover. We regularly use Veritas Firstwatch for failover. I have worked with Firstwatch and it is a really great product. Its been implemented in many large financial companies. I cant wait till this comes to Linux, it will be a big jump. Veritas Symmetrix allows you to have two separate machine both with the ability to fail over each other. The major benefit is that both servers can be working on a SEPARATE database so the machine is not wasted. We also use the Load balancing switches to do some fail over, this also works well but we havent done it many times. We also have some implementations of Oracle Parallel server, but these were a bitch to setup and they took a longtime,greater than 2 months!

    You have to think about the network, how your connected to their network, how big is your connection etc etc. Make sure you get the provider to give you a list of private and public peers that they have. They wont give you all the information all the time but push them and they eventually give you a lot. An NDA is usually all they want :). Make sure your bandwidth is measured as an average over the month. And make sure the provider does not make you commit to a CIR and BIR! This is not a good gauge of bandwith because they make you commit to a certain bandwidth charge and limit your burst rate. if you go over you pay a hefty premium.
    u

    The network is protected on the front end by Access Control Lists. These access control lists block all ports except for 80 and 443 for most customers. We implement a firewall on a separate Sun machine for customers who want more security. The problem with this is that it usually acts as a single point of failure for the entire site, with two webservers and one database server you can lose one server and still serve static content. With the firewall you lose one machine and your whole site is down! So we often load balance and fail-over the firwall boxes.

    As for backup we backup all the machines to a large backup array every night. We do over 700 machines a night for over 150 clients so it is very traffic intensive, but it saves changing tapes all the time,the robot does it :). We incrementally backup every night and then full backups every Sunday. Keeps all of our data nice and safe.


    You should definately go to the datacenter!! See what their engineers have to say, how the offices are set up etc etc. These companies must be the highest professionals, of course they dont have to dress like them they just have to act and talk professionally and knowledgably. Make sure they have a good disaster recovery plan and ask for papers on it.

    The one thing that you should definately remember and that is to Always get Service Level Agreements and Descriptions for every service they are performing or saying they offer. This is what they will go by in the contract so read em and get the truth about the service.

    When picking a provider there are lots of things to consider. There are so many of them now and there will be many more adding on, not all of them are the same and many are not good at all. Here is a list of some providers to look to and compare:
    Exodus
    Digex
    Navisite
    GLOBIX
    HarvardNet
    SureNet
    USI
    GTE
    The best providers are usually the ones with Managed services where they manage the O/S and web server for you. They leave the content and other apps up to the customer.

    I am putting together a page on choosing a provider so if anyone wants to help out with suggestions let me know at:

    johncrisp@hotmail.com

    Thanks!
  • If I was going to launch a new website that I thought the public might be interested in, I would start by colocating the box. I would not start by ordering a dedicated circuit specifically for the website, because I simply do not know how much bandwidth the website is going to need.

    If the site never took off, you will be out the thousands it cost to get the loop installed, the DSU, the router, and ISP costs.

    If the site REALLY takes off, your initial circuit might not be able to handle the load causing users to become frustrated and give up on going to your site.

    Only if I was delivering static content would I consider web hosting. For anything dynamic I like full control of my environment.
  • No, bad idea -- harmonics problems start creeping up when you have recursive backups. Get a halfway decent UPS to keep things up, and keep a generator handy. Even if a nuke knocks out the power grid, gasoline can keep you powered for a long time.

    Then again, if you have to run the generator for more than a few hours because of some catastrophic event, you've got bigger problems...
  • Agreed. I personally would never pay for banners on another site. But there is still demand for banner space (I guess clueless people who think banner ads work really well). If your site is good, people will go to you about buying ad space.
  • Make a site and browser plugin software that manages users bookmarks accross the web. I have 4 browsers each with their own bookmarks. The user logs into your "Portal" sees ads ($$$) and gets his/her bookmarks centrally located. As extra $$$, (and a privacy taboo) data mine the users bookmarks and sell the info for demographic moolah.

    Hedley
  • Check out Web Tools Review [photo.net] for a good sampling (albeit with a heavy AOLserver bias) of people doing the same thing you are.

    You WILL want to colocate. End of story.

    Use RAID 0+1 and learn how to set it up yourself in software if you're running a database.

    Don't run a database where it isn't necessary; they're slow. Replicate them where you do run them, and develop a real backup strategy.

    Run what works best for you -- Sun, FreeBSD, Linux, NT, whatever. Ignore the bigots, they're not the ones risking their financial futures on a venture. But don't get suckered by marketing.

    Favor security over convenience, but don't lose sight of what it is that's valuable enough to secure. In other words, don't go to either extreme. (it's very easy to forget this!)

    Good luck.

  • One of the best ways to learn about business in general, which you will need these skills for, is to examine other companies and see what is it about their "vision" that makes them sucessful. Carleton (Carly) Fiorina, the new CEO of HP had, in my opinion, a very good keynote address at comdex [zdnet.com] (I recommend you listen/watch it if you've got about an hour). According to her (and she's been throught more business education than most people I know, so I believe it's worth listening to), the next generation is bundling services to products - as much as web-based businesses would like to (or the one's I'd like to start anyway) have super-low overhead, and little need for research, I believe research is the biggest piece of a company.
  • 6.) hire another guy to do grunt work like backups etc. thay you've been doing till now.
    or think of a great way to automate the process, write some cool software to offload the grunt work to your silicon workers :)

    In a start up logistics is also a problem. Any code or tools you create to leverage your logistics will help. An automated database back tool is one. I know, 'cause this is exactly what I did.
  • by RobertGraham (28990) on Tuesday November 16, 1999 @05:51PM (#1527314) Homepage
    I work for a startup and help maintain the website. We currently get about 20,000 users/day, but are still growing rapidly. Here are my recommendations:

    It's about the content, stupid Ultimately, the growth and popularity of your site is determined by the quality of its content, not its looks. Don't worry about looks until much, much later. Too many people shift too many resources too early into good looks. Remember, you've got finite resources. I've seen many sites fail because they spend all their effort getting the look just right, and never get the content right. Marketing types fret a lot about protraying the right image and all that crap; they have to fret about something because they rarely understand any of the actual details.

    All code has bugs This simple law of programming applies to websites. Whatever ideas you have now about your website are not complete. Slashdot is always tweaking its content to create a better user experience. This actually dovetails with the point above: too many sites get customer feedback about things that need to be changed, but cannot because it would break the cool graphics, or the master design. Design your system NOW for constant tweaking, or you won't survive.

    KISS (Keep it Simple, Stupid) People get enamored with the latest technologies and build websites that require the latest browsers with all featuers turned on. If you do this, you'll kill your site. Test with Lynx and make sure it provides an adequate (though not wonderful) user experience, then it'll likely work for everyone else. Cool stuff like Java, plugins, scripting, etc. are nice for sprucing up sites a bit, but if you depend upon them, you'll kill your site.

    Focus! This is how marketing departments kill good work. Engineers try to create very focussed products that solve specific needs well, marketing tries to broaden the product's appeal, overloading it with features that end up satisfying no-one. Start broadening your appeal only after you've created a solid user base. For example, our company does business with another company that we have to help figure out very basic issues, because they aren't focusing on the technical problems but the "story" of where they are going. If they don't clean up their act, they will fail.

    RAIL Redundent Array of Inexpensive Links. Grab two DSL lines from independent providers; that's all you need for a really popular site. Of course, that's assuming that you've followed the KISS principle above -- a lot of sites have huge graphics that quickly eat up bandwidth. You could easily maintain 20,000 hits/day and not eat up a 384-SDSL link. The chief problem isn't bandwidth by reliability. Even hosting companies like AboveNet and Exodus go down when backhoes take out their backbones. A RAIL solution solves this problem: DSL lines are a lot less reliable, but two DSL lines (from different vendors) are more reliable.

    Backups You MUST have offsite backups. Also, assume any machine connected to the Internet will be corrupted (i.e. static content should be kept internally and regularly mirrored out onto the Internet servers).

    Hacking um, you WILL be "hacked". Plan for it. I mean it. For example, your servers front-ending your site will easily be hacked, but if you plan on that contigency, you can usually harden your database server against further incursion. Don't believe me? Follow these steps: (1) go to Yahoo and search for "wwwboard passwd". (2)about every other link will be a pointer to an WS_FTP.LOG, which you replace with the file "passwd.txt" (3) run these passwords you get through a 'crack' program (4) poof, you thousands of passwords with only a few hours worth of work. Note: in this example, firewalls don't help.

    Platform The underlying platform is irrelevent, in both security and performance. You should strongly consider PERL for dynamic content, only because it is the most used (and consequently, when you hire people to work on your site, this is what they'll know). Geeks like to fight over the most technologically elegant solution, but issues like hiring experienced programmers that can maintain it are far more relevent IRL.

    Manage growth You will be too optimistic about growth in the beginning, and too pessimistic at some later date. You'll do a bunch of stuff that you think will drive people to your site, but they will fizzle. Then out of the middle of nowhere something happens and hits shoot up 10 fold. Be ready for both (watch cash flow and don't overspend now, but be ready to upgrade capacity at a moment's notice).

    HITS Note that one of the Internet scams is people that promise to drive hits toward your site. This is all crap: all such techniques are publically available, and since this is your core business, you need to learn all of them yourself.

    Outsourcing Outsource everything that isn't your core business. This is the Internet baby, you don't have time to build a company. You can't hire people fast enough, and you can't hire good enough people. You also don't want to be giving stock options out to people that don't directly influence the companies growth. For example, don't have a human resources person doing health insurance, outsource it to a consultancy. Many of the .com startups use this approach and have surprisingly few people when they go public. Conversely, the previous section is a good example: when it is your core business, DON'T hire consultants or outsource it -- do it yourself.

    There's more, but I think this message is getting long enough.

  • If you're willing to do a bit of a stakeout, a good domain is easier to get than you might realize. From the domain-hunting I've done, it seems to me that the vast majority of registered domains are owned by speculators who submitted an email template with no intent of paying for the domain if they can't resell it. And almost all of these are deleted within a few months for non-payment.

    From what I've seen, I believe that almost all (maybe even 80-90%?) "good domains" (e.g. common words, or 2-4 letters long) fall under this description.

    Now that it's more difficult to register without paying, perhaps a great many new domains will become available. As far as I've seen, though, these domains are always registered by somebody else within hours or minutes by someone else.

    Disclaimer: this is a theory, based on observation but not proven by empirical testing/sampling.
    -
    <SIG>
    "I am not trying to prove that I am right... I am only trying to find out whether." -Bertolt Brecht

  • My list of suggestions:
    1. In the begining co-locate. Not only is it less headaches but small tasks like DNS etc are handled for you. Plus if you chose a co-locate carefully you will get a tremendous knowledge base you can draw from. Also if you get a hacker you will have tons of help smashing the fool.

    2. backup on and off site! The offsite provides you security period. The onsite means you can have a setup where two drives are available for booting (using a sync software package) and if it drops just have the guys working at the co-locate reboot you into the other drive.

    3. Print contact info, reboot instructions etc out and paste it permanently on your server. This makes it easier for people to contact you or assist you if needed.

    4. If your going to run Unix or Linux make sure the service isn't an NT house. That way you don't have to explain the 3 finger salute 20 times to the tech on the phone.

    5. CONCETRATE on the quality of your html the headaches you will save by doing it right the first time are imense.

    6. Don't waste money on gee wizz and wizz bang. A 4.3 gig hdd is probably 10 times larger than you will need. 64 megs of ram should do in the beginning but 128 is better. Make sure all the fans are ball bearing (it'll save you in both maintenace time and in burned up cpu's) In fact extra case fans are a plus.

    7. Don't waste money on a monitor or a keyboard at a co-locate. Most of the ones worth a dang have "crash carts" with monitor and keyboard.

    8. Unless you plan on doing a lot of onsite installs don't waste money on a floppy drive or cdrom drive. If you need them just carry an old one with you.

    9. Make sure your OS you choose is remote configurable and maintainable. It'll save you both time and gas money if you can fix small problems from wherever you are.

    10. Pay attention to your log files. Not only are they a wealth of info but if the dang things get too big your system goes gaga (great technical term isn't it) Develop the habit of daily checking/downloading/deleting the files.

    11. Blanket banner adds are a waste however trading links and customer targeted advertising still bring in a large amount of viewers.

    12. Don't skimp on motherboards fans powersupply, ethernet, cpu and ram. Do skimp on vga, don't buy sound, cdrom or floppy.

    13. Find out the amount of space you will have in inches BEFORE you buy the case. Nothing like buying a 10 inch wide case and finding out you will have 8 inches of space.

    14 Install a 10/100 meg ethernet card that way if your connection is to a 10 meg lan and you get upgraded to 100 you don't have to take down the site.

    15. Remember having a good website is like having a happy baby. The more attention you give it the better it will turn out.

    Hopes this helps.
  • I don't see the problem with software raid at all. Software based raid such as ccd and vinum work very well. Your controller can fail too (or even have a buggy DSP), what's your point?

    Don't get me wrong though; I work almost exclusively with DPT fibre channel appliances with multiple servers connected. I just don't see your point that software raid isn't fault tolerant. There's a possibility of such raid data storage appliances failing as well.
    ----------
  • I've found pair very handy while I got started, but I'm going to outgrow 'em around january, and their high-end QuickServe package unfortunately inherits a lot of the policy limitations [pair.com] of their shared servers. On shared servers, they make sense.

    Pair was a useful and pretty inexpensive service before I was committed, and it was great to have access to SSL and MySQL without having to sweat the details, but my next setup looks to be a ppc-linux system at above.net.

    bumppo

  • by goon (2774) <`ua.ten.ecapsten' `ta' `liamnoog'> on Tuesday November 16, 1999 @06:47PM (#1527332) Homepage Journal
    Here's some laws I've formulated after working in 3 startups from '96 to now (sausage.com, www.ringtail.com.au and the company I work for now). They relate a lot to ASP (application service providers) but could be applied to binary apps companies.
    1. hire less, hire smarter
      - company iq==(total employee iq/employee#)
      own the data
      - owning/possesing the data allows you to do lots of things with it. Data hosting is a core activity of ASP's. Own/posses the data, you can do lot's of neat things with it.
      complete the data loop
      - from customer/user input into a database, useful information is filtered back to the client using the Internet and it's protocols.
      use the internet protocols 2 your advantage
      • smtp:
        - email back results of non-immediate jobs
        ftp:
        - allow large files/data that can't be email to be accessed.
        http:
        - presentation layer of services
        xml:
        - new one but such organisation could revolutionise your text file storage.
        extend and leverage your logistics.- by using the internet and your database, development tools to automate functionality: ie: web backups of databases the transfer DDL, BCP data, codebase, log files from different areas on the site to a centrally located removable hard disk or Jaz disk: I've don this myself.
      develop a minimal rock solid product.
      - one that does not crash or crashes minimally. Customers wont praise you but will let you know if it does not work.
      give your customers a cheap basic product to start with.
      - dont give a cost hurdles for customers to adopt your product.
      leveredege existing binary products
      to use Internet facilities with objective of upgrading customers to the web.- migrate those binary products to use internet protocols with the idea of upgrading all/most of your services to the web.
      Become your own ISP
      with control of web servers, DNS, databases ISDN internet link etc.- you have full control of your domain, you can do everything you possibly can.
      Maximum bandwidth decides you maximum audience.
      - bandwidth dictates the speed, reliability and user experience: Use slashdot growth as an example.
      Fund initial expansion without going into debt.
      - dont waste money and go into debt. Finance growth on profit until venture capital is possible.
      Have bloody lots of fun
      - if you cant code what you want, play the games or build robots, code some great code you wanted to do you will not enjoy the work. Play hard, work hard.
      Dont give up
      - startups fail, dont let this stop you. Check out what happened to Crack.com. Do you think these guys are giving up. Wonder how long ddt and jt are back building FPS or RPG post Transmeta and JiJit? failure makes you smarter!, comebacks make you legends :)
      try to use a higher level dev languge
      , use the source!: scripting not binary- binaries are hard work. The develop, compile, build, create install, release cycle is way to long. Scripts are the way to go. Error, just go to the script (text file) and make a change.

  • I really can't agree with the ideas of hosting it yourself/building a network for it.



    It's always been geek fantasy to have the T1 to the house. Fantasy is rarely practical.



    It will cost you in the neighborhood of $1500-$2000 a month for a t1.

    A t1 really is not that much bandwidth. I've worked for a few companies that do webhosting like iuinc.com (I don't work for them now, just host servers there). Hosting with a mid-sized hosting company is a much better solution.



    The problem with doing it from your home/business means you have to have somewhat of a datacenter in the building. This requires a lot of Hardware :



    • CSU/DSU
    • Redundant CSU/DSU
    • Router
    • Redundant Router
    • UPS
    • Ethernet cable
    • Link to the net
    • Redundant feed
    • remote-access switch and phonelines.

    This is all very costly.. You also have to put
    into account the money & time it will take you to purchase, configure, test, and put into production, all of this hardware. Then maintenance
    is a big hassle.. maintaining the systems is one things.. but rebooting @ 2am when you're
    away on vacation? dealing with the telco/bandwidth provider when something goes wrong.. a web-hosting company
    charges a small premium basically for hosting, and saves you a LOT of $$..



    It is much more effective to host your server at a company who solely does that. You can expect to pay $200+ for that, plus between $8-$15 per gigabyte of transfer. That's great, if you only transfer 30 gigs a month.. that's $500 a month.. compared to $2500-3000 for your t1 & redundant t1 + hardware costs of doing it yourself.. money better spent on advertising, site development, video games..




    I'm currently in the process of doing the same thing, setting up a small hosting company.. to deal with people I meet @ places I work with (a good point to getting business like this is people you meet, interact with.. at clubs, bars, computer stores, libraries, etc).. know less than a ny times journalist about the internet.. and for your well-developed server, and highly knowledgable self, they'll gladly pay $50-$100 a month for basic hosting.. since you can provide for them a lot more than a large hosting company can. I've got a friend @ sony music who does web/e-commerce design on the side.. and got me to do it because he gets 2-3 clients a week asking him for a place to host.. and refers them somewhere else @ $70 a month for basic hosting..



    I'm finding it a better idea to build two servers, setup a good tape backup system, connect them to a raid (i'd like to go nas.. but it's not cost effective yet).. and write some good easy user-management cgi's in perl for my customers to do their own configurations.. so i'll just be sitting back, getting the occasional new customer.... and hopefully soon be paying for a t1 to my house (I am a geek) that I won't have to share

  • by dennisp (66527) on Tuesday November 16, 1999 @09:12PM (#1527349)
    "Focus! ..."

    It would be smarter for the company to sit down first and discuss possibilities. Market research is very useful. Your problem doesn't lie in the fact that "engineers" supposedly know all and marketing doesn't know the web from their ass -- it's that politics and lack of trust are getting in the way of useful group input. A lot of expert opinion definitely lies outside the domain of the programmer/sysadmin/html design artist (have no idea why you claim these people are engineers).

    "RAIL ..."

    I don't see your point. Any respectable web provider will have a number of redundant links. Providers such as exodus have the ability to re-route traffic within minutes (even seconds) of reported problems. They are much better equipped to deal with such things than you. Note that most DSL providers won't let you have AS authority over a group of ip addresses, so redundant connections are completely worthless because you can't re-route traffic when one link goes down. If the company absolutely needs their bandwidth in house at inception, then they should get a t1 with a shadow t1 connection from another provider for redundancy.

    "Hacking um, you WILL be 'hacked'."

    Oh give me a break. Just delete all default cgi and test scripts and make sure you're programming securely. When and if you get hacked, just restore from backup.

    384 sdsl line is NOT capable of handling this many hits. Traffic spikes will come, especially if you are linked or were recently visited by a search engine spider (and have certain popular key words on your page).

    As for simplicity, I completely agree.

    Content on the other hand. Yes, content is central to a sites success. Bad site design will prevent that content from being seen or turn users away. It is a very important element to take into consideration. While I agree that resources and time are limited, there are many established sites that you can use to get inspired.

    to the person who 'asked slashdot':

    A much cheaper solution is to start your site out on any number of shared hosting providers such as pair networks [pair.com] who are very cheap and support mysql, php, cgi etc. There you can get 12gb transfer a month for only 27 bucks (or 6 gb transfer for 15) with php, mysql, cgi and a telnet login to one of their hundreds of freebsd boxes. I went from 12 to 50 to 100 gb a month. I'm sure there are hundreds of other similar providers. It's smarter to do this first than to make a very big investment and get a dedicated connection and whatever servers. Let others provide these services for you while you build a customer/client base and eventually you will outgrow the price curves of companies such as these. Remember, if your site has any downloads, you'll definitely need available burst speeds that these providers have. Slower static links just won't cut it (well unless you can compromise and put every large file on medium quality xoom.com or whatever links).

    If your idea is really that good, then I suggest you scout out possible competitors and siphon whatever useful data you can. First identify your target market and the viability of your business model. Are you going to try and sell something online or rely on advertisements? Is this market already clogged with other hopefuls and/or will they soon join in on your good idea? How will you differentiate? If your site will be complex, are you standardizing the design and implementation to prevent many wasted hours? Are you establishing relationships with people in that particular industry? How will you promote the site during and soon after your go live date?

    For one, it's very important to standardize your design and try to modularize your code. You don't want to know how many companies I've consulted for and who decide to change their entire flawed backend model after building months of content. It's not fun.

    Remember, competition will come if your idea is potentially profitable. You want to build up a good repeat client base before others do.
    ----------
  • by dennisp (66527)
    "Become your own ISP"

    Yow. It's true that access to your data is important, but one of the main fallacies of ASP's is that they are slow. You want the customers to have an as fast as possible experience. Providers such as exodus and level3 offer very good data center services including backups, closeness to the end user since they can set you up with round robin dns using more than one data center, their expertise in dealing with 100's if not 1000's of companies (who likely need similar solutions), redundancy and reliability (can't stress that enough for the ASP market). You seem to get it with "Maximum bandwidth decides you maximum audience."; so I guess you're just trying to tailor your post to the person who asked slashdot (though I doubt it had anything to do with ASP's :).

    Instead get a t1 (or other decent dsl/wireless/whatever) connection so that you can update your data/files/databases/assets faster. You don't have to be in physical control of the hard drives to have actual control over the servers. Just make sure the data center is close by -- otherwise you'll be paying huge service fees to these companies.

    "develop a minimal rock solid product"

    And how will that help you differentiate your product from the other 3000 so called revolutionary ASP hopefuls? There is so much room for improvement in the ASP market, it's scary.

    "leveredege existing binary products"

    This is the main problem faced. The web browser is often terrible and slow and doesn't exactly enable this server side computing paradigm. There are a limited number of products at this time that you can pattern yourself after. It's better to look at the needs of a certain industry or the similarities between a group of geographically separated company/companies. Take into account the current technologies they are using and integrate them. Don't try and emulate them unless you can do it just as well or better. Example: It's better to just send word or pdf files than to concoct some online format. Just include features that will bring them closer such as comment and message board systems for revision and peer review.

    Some open industries for such include:

    a) general family and friends networks. This includes file sharing, message boards, photo galleries, electronic greetings, possible calendaring with events/birthdays/special occasions and possible e-commerce tie-ins. News headlines are usually included as an extra spice. Examples: familypoint and visto

    b) distribution networks

    communication is key here, as is numbers, schedules, private and secure messages, and strategy planning.

    c) geographically separated offices

    Just about every local communication plus to touch base, actually feel like a community (often not possible over phone and e-mail). This is treading into lotus territory but these kind of applications can be easily modified to suit a particular business model (instead of the company modifying practices to effectively use notes). Contract or remote developer to developer groups or developer to q&a relationships (bug tracking, design issues, ui, specifications, dates, etc etc)....

    c) I'm bored, won't finish.

    Specialization is especially needed. Law, accounting, investment and banks, retail sales and advertising firms serve as some examples.

    If you're an up and coming ASP, then make sure to try and circumvent the obvious failings of centralized, typically slow services. You may be in the wrong market altogether. Solutions with hardware tie ins may be many times more appealing to medium to large sized companies.

    "Fund initial expansion without going into debt"

    Easy to say, hard to do. Without a tremendous idea or a reputation to get funding, there are some major barriers to entry (bandwidth, support, contracts, research).

    "try to use a higher level dev languge"

    Agreed. Optimize for speed as per forecasted need. Embed c in perl or create ISAPI or apache c modules. Be sure to optimize database transactions and often used dynamic content includes.

    "complete the data loop"

    Oh my god, someone who understands. User input and needs are very important. This is especially true for established businesses. If you're not willing to meld your process to meet their needs they are just going to go elsewhere.

    xml; Is interesting, though similar things can be done with server side interpreters such as perl, php, python/zope and/or corba, whatever.

    Of course, all these statements are incredibly obvious and equally brain dead easy for any knowledgeable developer to pull off. The hard parts are in the market research, establishing industry relationships and input to meet their needs, funding, pr/hype, and luck.

    Don't worry, I'll slap myself silly for ranting.

    ----------
  • Its definitely not the banwidth. I have some corporate dedicated servers serving image/video/audio assets here at dn and they are always incredibly fast. I even get ~20ms from home (due to their uunet oc3 +1 t3). Check out their connectivity [http]. I only ever ever go through their teleglobe, qwest and uunet connections as dictated by bgp routing tables, so I can't be absolutely sure though.

    Slashdot seems fast enough for me. 100+ comment pages load in less than 5 seconds (remember they do extensive database queries). There are some days when the site seems to be down for 5-10 minutes. Actually, I've tracerouted on occasion and found that the box was up but the httpd wasn't responding for some reason.

    As for server hosting centers though, I haven't found anyone better than dn when it comes to multiple connections to just about every major bandwidth provider in north america. MAE's and notoriously oversold sprint-naps and other public exchanges are completely avoided 98% of the time.

    You can't go wrong when they're likely to be directly the same backbone used by you and your clients isp's.

    I wonder if employees get extensive free services. Drool.
    ----------
  • There are a lot of good tips in other posts, so I'll throw in my 0.2 cents worth. (The effects of inflation...)

    • Keep It Simple. (Forget the Stupid part - that gets in the way of you doing the keeping it simple.)
    • Memory, not processor power, is the key. My Masters was in web-based multimedia delivery, and a low-power Pentium with a decent amount of memory was blitzing everything else the University had.
    • Seperate disks for data and software is a MUST! This not only improves performance, but safeguards half your site if one of the disks crash.
    • DIFFERENT PRESENTATION FOR DIFFERENT PEOPLE! This is a -BIG- result my research proved. Lynx users -won't- be able to get the same l&f as someone using Netscape or IE. Modem users -won't- want to download megabyte images, T1 users might. Use templates, dynamic pages and some sort of flow control to determine what people get.
    • English is not the only language! If you can get someone to interpret pages for you, do so. Use multiviews to select which language to display.
    • If you have any bulky, static content, put that on a seperate server and cache it! Squid is ideal for this.
    • Do a little market research. See what the demand is, for the information you're wanting to deliver. If it's a totally new idea, this might prove difficult, but it's worth it. Estimate that 10% of those who say they're interested in such a site might actually visit it. Once you can guess the size of audience, you can guess the size of link you need.
    • BACKUP, BACKUP, BACKUP!!! I can't tell you how important this is! It's VITAL! Hard disks can go bad without warning, and the one thing you don't want is to hear all your hard work go down the drain with that terrible grinding sound. I've heard it too often to be complacent.
    • If your market is such that reliability is essential, use High Availability software. There's plenty of this for Linux, now.
    • Journaling filesystems may be a good idea. Reiserfs not only is faster than e2fs, but has journalling fully suported, now. It might be worth a look.
    • UPS! Not the postal service, but the power supply. You -don't- want your system going down, in a spectacular nose-dive, just as your top sponsor pays you a visit, in person or on-line.
  • Not to answer another persons question to another person... ah hell I'll do it anyway.

    Software RAID is bad when your system goes down and you have unwritten data left on the system, normally waiting on I/O, CPU cycles, etc. Oracle is a big one on this, it's very bad when you have a thousand transactions commited but waiting to be written out and the CPU fails or someone unplugged the box, or whatever.... you are just S.O.L. at that point, all the transactions are gone. With hardware raid, any "decent" raid controller is redundant at the cache level also with battery backup, so all those transactions WILL get written out; I see that as the biggest beef with software raid (not mentioning speed, cacheing, CPU cycles, multiple hosts, FCAL, etc.) Software RAID is meant as a solution when you want raid, but you can't afford to go hardware (make no beef, hardware raid is damn expensive).

    My opinion:

    You should never skimp out on hardware raid, and go with software if you are able (can afford) to, that extra budget slush money going to your workstation isn't as important as the lost data waiting on the your servers bus when it goes down.
  • Even still, the system shuts down and the transaction will likely not finish. Granted, it's better having a dedicated 32 - 64+ meg cache, but how often is this going to be a case of a single built up transaction *already* completely in i/o cache. It's also likely that the computer will fail during such a transfer.

    imo, the hardware solution is better, but the software solution can still be viable. Hardware and software solutions aside, you should already be running multiple redundant and identical databases (as well as weekly and monthly backups [daily if possible]).
    ----------

The degree of technical confidence is inversely proportional to the level of management.

Working...