Where Do You Shop for Server Components? 447
Devi0s asks: "Along with many other Slashdot readers, I have been building my own PCs for years. I use hardware review sites such as Ars Technica, Tech Report, and Tom's Hardware Guide to research the components and pick out the best, and I use PriceWatch and ResellerRatings to find the best deals and to make sure I am dealing with a reputable vendor. I work in a small consulting firm where money is tight, and I'd like to test the waters with a few ideas of my own. In each case, various servers and external storage enclosures are needed on the cheap that will be pushed to their limits. Are Slashdot readers building their own servers and storage enclosures? What web sites provide the latest news, research, and and comparisons for server hardware? Where do you go to buy server components and vet your vendor?"
IT "Pro's" dont build servers and storage devices (Score:5, Insightful)
"...various servers and external storage enclosures are needed on the cheap that will be pushed to their limits."
You are about to discover why server vendors are still in business, even though commodity parts just reach new lows in pricing.
I wont bust your balls, or tits(?) over trying to do IT on the cheap.
However, you will need to make some choices.
Cheaper hardware will only buy you hardware designed for consumers ( do you recall the IBM Deskstar models that had a monthly hour limit of usage ? ).
If you need to buy hardware that wont blow up under load, and you can get replacement parts for ( especially outside of business hours ) you should stop and go back and review products from IBM, Dell, and Compaq.
Recognize these vendors call it a server since they do test these things under load, test compatibility under cirumstances that your describing, and provide service so that your consulting shop wont be twiddling their thumbs waiting for you to run down to the local swap shop to get a new motherboard.
Those of us responsible for maintaining services ( DB, Email, etc ) dont build servers unless our backs are against the wall. Even then, we buy HW from the same vendors who make the servers.
Why ? Our job isnt to build hardware. Its to make email flow reliably, keep end users data available whenever management is willing to pay for it, hopefully you get the drift.
Since your in a small consulting shop, the big goals for the shop is growing clientele. They money will be well spent, when you and whomever else is responsible for the backoffice equipment ( in a small group everyone wears lots of hats ) spends your time building the customer base.
There will be lots of folks here who will say, sure go do this, this and this.
Hopefuly, a few will try to influence you as I have, and suggest you use a Cisco grade product, versus Linksys.
Re:Cost analysis (Score:5, Insightful)
Of course, lots of people take the roll-your-own approach just because that's the only way they know. The last regular job I had was for an internet services company that had started out in the owner's garage ten years earlier. For our virtual web host business, we still used the RYO server management software the owner had written back in that garage! And even though we were now managing a data center with thousands of systems, everything in sight -- the phone system, the customer support database, the procedures we used to checkin hands-on customers, even the tests used to screen potential employees -- everything was very do-it-yourself. Not the most cost-effective way to run a business, but the owner simply knew no other way to get things done.
After all we wouldn't be techies if we didn't enjoy playing with technology.
Re:Fry's and Best Buy (Score:3, Insightful)
HP + 3yr lease == happy clients. (Score:3, Insightful)
Our job is to keep the services serving.
So we get equipment that is designed for that purpose, with overnight delivery of replacement parts and 24 hour tech support.
I prefer leasing servers for 3 years because that's usually how long the various fans, power supplies and disks will last.
Losing a server in the middle of the day is just about one of the worst things that can happen for a consultant.
Your client loses all their work/data since their last backup.
So, you take every precaution you can to ensure that that will not happen. And leasing the machines means a small monthly payment and brand new machines in 3 years (which you will be paid to install/migrate to).
Don't ever risk your client's work or data.
Re:Strictly cash and carry for me... (Score:1, Insightful)
Re:Newegg (Score:5, Insightful)
Google it if you want to know more.
No, you provide the link. Prove your accusation or you're nothing better than a common troll.
It's not worth it (Score:4, Insightful)
The biggest reason is that if you buy a server from one of these guys, you know that you're buying a reliable, fairly stable platform that will be supported throughout it's useful life. I personally use all HP servers, and if I have a hardware failure I know that I can make a 10 minute phone call and have an engineer onsite with a replacement part in 4 hours. If I roll my own, I may have to do some troubleshooting with a number of different component manufacturers, especially if there's any finger pointing going on. With HP (or any other big manufacturer), there's one number to call and no fingerpointing.
Also, the big three vendors tend to be better able to do troubleshooting than a number of smaller vendors. I know that HP includes software for management and monitoring with their servers. I have been saved from major downtime on multiple occasions because Insight Manager has detected an imminent failure and I was able to preemptively replace the failing component. So in that sense they are also better supported than a roll-your-own solution. When you're talking about storage systems, EMC goes even further with their call home monitoring/support system.
Finally, the major vendors all perform compatibility testing on their platforms. If you buy a server pre-built from any of the big three, you can generally choose from a number of different flavors of Windows, UNIX, or Linux and know that the server is compatible and supported on that OS. If HP releases a new driver for a fibre channel HBA you know that it won't cause problems with their SCSI HBAs. If you roll your own, you have to do all of the footwork and compatibility testing yourself.
If I need a new server, I have basically two options:
1. Go to a big three vendor, spend 10 minutes configuring it online, and place the order. I can have the server shipped to me and even receive it the next day if I want. All I have to do is unpack it, rack it, and install my applications.
2. Shop around for parts, comparing reviews and prices, and hopefully come up with a config that I can use. Then order the parts from one or more vendors. Then wait for all of the parts to arrive, because the odds of everything shipping and arriving on the same day are pretty slim. When everything arrives I have to make sure that I have all of the drivers available, and then assemble the server. Then install the OS and necessary drivers, hoping that there aren't any compatibility or configuration problems to be solved. Finally I can install apps and rack it.
In business, time is money. Option one takes less than an hour of my time from placing the order to server up. Option one also is far less likely to cause me to lose uptime or cause other headaches from a support standpoint. Option two may save me a couple bucks upfront, but in the long run it's going to cost me far more. Option two is especially bad if a part fails and that particular model is no longer manufactured.
There's a popular saying in the computer industry that goes "Nobody ever got fired for buying IBM." There's a reason that it's true, and it extends to the other big manufacturers as well.
Re:Only Newegg (Score:3, Insightful)
Newegg has been great for me as far as personal purchases go, but they are unsuitable for my company and many other business because of their unwillingness to accept a Purchase Order in any manner. While I am willing to use a credit card for personal purchases I am not willing to do business with a company that will not extend credit (NET 30, etc) to any other company on the planet.
Re:Cost analysis (Score:3, Insightful)
The other major thing to consider, is that different architectures handle I/O better than others. For I/O-bound apps, stick with Opteron systems, and don't skimp on the motherboards. Apple's XServes would probably be the next best, and Intel-based servers at the bottom. AMD's architecture simply can't be beat for I/O, as you will have essentially separate busses for memory and a few I/O busses (unless some are used to link to another CPU).
I've had good luck with 8aNet [8anet.com] for servers, NewEgg [newegg.com] for networking components, and basically just whichever reputable name has the best price on pricewatch for disks.
Re:Only Newegg (Score:5, Insightful)
Re:Cost analysis (Score:3, Insightful)
All of the commands that I've used in Linux work on the Mac. They have the full suite of gnu apps that everyone knows and loves.
If you don't like darwin ports, then try fink, which is a debian package system. Most things compile from source though, and many of the major vendors provide binaries for the Mac.
The Dells and HP computers in the same price range only come with a single power supply as well. I've never been a fan of huge computers with massive redudancy anyway. If it's that critical, have a completely different system that will provide the same services should the primary server go down. Or just expect it to go down once in a blue moon. Apple does provide you with spare parts if you get the expensive support option.
supermicro (Score:3, Insightful)
1) Shipping problems would always come up where they wouldn't send out items until like a week after I bought them.
2) I recieved quite a few parts that were DOA
3) Putting some of the 1U's together was a huge pain, especially trying to find 1U cpu fans for the faster processors
4) hardware would fail, and due to it being about a year and a half since I built the machine, and hardware changing so rapidly recently, It would be hard to find a local store with replacement parts.
I think out of the 9 rackmount servers I built, I've had 5 motherboards fail, 6 sticks of ram, and a power supply (not counting the stuff that was DOA). Most of the failed hardware I would say happened within the first week of the server being online, so I dont know if that should be considered DOA too, but i'm not including it. Anyways I've recently been buying Supermicro [supermicro.com] servers and couldn't be happier. All the machines I've got from them, and that my friend who recommended them has got from them have been running great, and they're much cheaper than IBM or any other big manufacturer (though if I had the funds I'd rather buy IBM). Anyways, before I quit building my own machines I found a great place to buy rackmount cases which is rackmountmaster [rackmountmaster.com]. All the cases I've got from them are laid out great, have good air ventelation, and aren't rediculously priced like so many other rackmount manufacturers.
Re:It's not worth it (Score:4, Insightful)
Server? You mean, like the tens of thousands of machines Google runs? The tens of thousands of commodity-PC-based machines? That they don't even bother swapping out when they fail, until the next regular maintenance cycle?
I understand the people who say "buy Dell or HP or IBM", and I understand the people who say "buy cheap and buy three of them for half the price of Dell/HP/IBM". But it all depends on what you need. If 24 hours downtime will put you out of business, you'd better go with Dell/HP/IBM and buy a spare as a hot backup. If you can live with a few days of downtime, go for high-end commodity parts and buy spares of everything (and 99% of the time, you'll have less downtime by swapping out a drive/motherboard/whatever than it takes just to get the Big Boys on the phone - But I acknowledge that, for some uses, that remaining 1% makes paying almost any amount worth the peace of mind it brings).
Incidentally, the idea of "commodity" hardware doesn't necessarily mean anything by comparison with "server" hardware - Yes, Dell guarantees all the parts work well together, and will replace the whole unit via overnight shipping if it fails; but at the heart of every Dell server, you'll find nothing more complicated than high-end commodity PC parts. Parts such that, given a list of them, you could build it yourself off-the-shelf from Pricewatch for $5k vs $20k+.
Buy a brand name server - here's why (Score:1, Insightful)
Resilience
Redundancy
Compatibility
If a PSU fails do you have to down the server, no, because you have a server with hot plug redundant PSUs, slide the dead one out and slide a new one in.
If a disk fails do you have to down the server, no, you pull out the hot plug SCSI drive and slide in a new one and let the array rebuild.
If an on board fan fails do you have to down the server, no, you slide the server out of the rack and pop off the top or if floorstanding you remove the side, unplug the failed fan and plug in a new one.
If an array controller fails what do you replace it with, will it detect your disk array? Will it blow away your configuration? My experience is with HP/Compaq, their array controllers are backward compatible, plug in a fresh one, connect the arrays and your original array is back up. With Dell they have different manufacturers of the chip sets on the array controllers so a little more thought is required but the same should apply. With a home brew server you really are going to have to put a lot of effort in to make sure you get the right controller so you can see your data, do you really need that much extra work when under pressure? Is your hourly rate that low that you can't afford a brand name server?
SCSI versus SATA
SCSI disks have a duty cycle of at least 90%, the latest and greatest SATA disks have a duty cycle of 60%, that means that the typical amount of usage per 24 hour period will not exceed those %ages and those disks are guaranteed and sold on that basis, if the disks are used above those figures the lifespan of the disk will be diminished. SATA may be approaching SCSI in many ways but it isn't there yet so for serious server use unless you have multiple redundancy in your raid arrays then it is less than ideal.
The cost of a new server is so low now that in the professional environment it really isn't worth cobbling together some collection of parts. If you are running a test server in your bedroom then maybe.
And lastly, pay for maintenance, let someone else deal with the headache of changing a failed mainboard, the cost of maintenace on these modern boxes is so low you would be foolish not to.
Re:It's not worth it (Score:2, Insightful)
Poor counterexample. It's rather rare at the enterprise IT level to have 10k boxes running exactly the same application with a ridiculously easy fail-over scenario. Moreover, I seriously doubt that Google runs their transaction based stuff (e.g. ad signup/payment and DBs) on commodity PCs.
Parts such that, given a list of them, you could build it yourself off-the-shelf from Pricewatch for $5k vs $20k+.
The machines are the same only if you think high-availabilty features, engineering quality, and extensive compatibilty testing are worth nothing. (Dell excluded; they probably do use commodity motherboards.) As you rightfully point out, for non-enterprise applications, mostly they do happen to be worth nothing. But that in no way makes the two machines equivalent.
Re:Cost analysis (Score:3, Insightful)
Have you ever used another UNIX-like system where adduser works the same way as on Linux? Take a look at the UNIX Rosetta Stone [bhami.com] for a comparison of the different commands for doing this kind of thing. From your earlier post, you complain that they didn't just put a nice GUI on FreeBSD (completely ignoring the fact that OS X is not FreeBSD, it's OpenStep with some of the old 4.4BSD stuff replaced with FreeBSD stuff). Perhaps you've tried running adduser on FreeBSD? It has completely different options and syntax to the Linux command of the same name.
Re:Cost analysis (Score:2, Insightful)
Wrong. It is paid for as an overhead cost. Think of it as all the time you spend researching, ordering and integrating components into a system vs. the majority of that time being spent out selling/building/conducting your business.
Time is money. Ask those guys with private jets (owned full-out or in fractional ownership) why they spend the money on them. The answer will be because the efficient use of their time (versus the two hour-plus wait at the airport, inability to work on the plane, layovers, etc) is far more valuable than flying commerical.
That said, sometimes for special projects with special needs, ROY is the best solution.
It all depends upon how you pay for it.