Catch up on stories from the past week (and beyond) at the Slashdot story archive

 



Forgot your password?
typodupeerror
×
The Internet

What Setup Is Best For Large Mailing Lists? 19

Super_Frosty asks: "I run a large mailing list that has expanded beyond my means. Perhaps you have heard of it - its headquarters is at this site. I currently have about 10,000 subscribers, which I am running from a small group of sad Macintoshes. I can hardly send out the daily message anymore. Anyhow, I can't find anybody who knows how to administer a list this large. I am beginning to have embarassingly close calls with people flooding and spamming the list due to security problems. My hardware can't handle it, and most software packages aren't meant for it, either. Can anybody reccommend a hardware and software solution for this list? Surely, some Slashdotter has administered a large list - I hope!"
This discussion has been archived. No new comments can be posted.

What Setup is Best for Large Mailing Lists?

Comments Filter:
  • I just noticed this post. Can you provide some information on what you are currently using? I currently run a list with between 700 and 800 subscribers (mail.liberal-judaism, do a search for the URL). We've been as high as 1000. I used to use a hand-grown perl script approach; I now use the listproc based at shamash to run the list and it seems to be working just fine (this is the latest version of listproc). In any case, some more information would be useful. Daniel
  • by Matts ( 1628 ) on Sunday May 14, 2000 @05:28AM (#1074105) Homepage
    The only mailing lists I'm on that have near instant response time are running ezmlm with qmail (including some large lists like the mod_perl list and the perl5-porters list). I can't recommend this setup enough, I use it for a very low traffic mailing list for AxKit, but besides the low traffic, administration, albeit by command line utilities, is a breeze (there are web admin tools available, and admin by email is possible too).

    Every other list I'm on, run on a variety of setups (the most common appears to be sendmail + majordomo), has really poor latency of around a couple of hours. For a daily email that might not be so bad, except to realise that the latency transfers directly into list delivery speed.

    There's not much more to say about it. Try it - I never regret it.
  • From the look of the URL you gave us, it seems that this isn't a real list like most people are thinking... It looks like a daily message from one source sent out to a bunch of people, without any feedback/discussion from the subscribers. If this is true, qmail alone can handle it, using some custom perl and qmail's dot-qmail file format. Using this method, the distribution can easily be split amongst several servers.

    If the subscribers do need to have discussions, then I would suggest qmail + ezmlm. If I recall, it too allows lists to be split amongst many locations, albeit with more trouble.

    Now, none of these suggestions are going to help if the problem is bandwith, or anything other than server speed/power/etc. What exactly seems to be the bottleneck?

  • by Anonymous Coward
    Set you up a little linux box with a moderately tweaked sendmail setup (at least queue time and retries need to be non-default). Get procmail and SmartList [procmail.org]. its more customizable and stable than the other options mentioned, I think, but it will be a bit harder to set up than some of the others.

    I use basically the same software to do a job very similar to what you're doing, just not on a daily basis... more like every 2 or 3 days. I've only got 3Kbps of bandwidth available to it; and it finishes 10K emails to lists of 10,000 to 15,000 addresses in under 12 hours, usually.
  • As long as you choose reasonably decent hardware and a good OS (FreeBSD or Linux seem like good choices for this), you should be OK. Lots of memory (128M+) would probably be useful, though.

    Mailman has a great user and admin interface (all web based, very intuitive to use), but it gets kind of hard to admin once the mailing list reaches any size at all (it's kind of a pain to work with only 150 members on one list I admin). I don't know much about MTAs; sendmail seems to work OK but I've heard that postfix or qmail are faster. YMMV.
  • by Anonymous Coward
    I'm posting anonymously because I can't log in... not sure what the problem is.

    I built a large scale mail cluster for our company. We send mail to ~1M addresses daily.

    MTA doesn't really matter that much, I've found; people will argue with me here, but sendmail is fine, and I use it becuase I know it best. What is important for bulk outgoing mail is an optimized injection agent - the beast that feeds the MTA (sendmail or what have you). Initially, we used bulk_mailer, but found it didn't scale well for really big lists - it would choke a machine. What we ended up doing was writing a limited MTA in perl that talked to a list of mail servers, sending one message to a mass of people (still tuning, but about 1000 seems right) - 1000 RCPT to: lines, one message - then switching to a random mail server and repeating. This might not be clear - we have a client/server cluster, one master sends mail through a lot of slaves. We do several different lists, so the master has many instances of the custom script running, each picking a slave at random from which to send mail.
    Slaves can fall over and they'll just be skipped.

    The master is still a single point of failure. I'm hoping to modify things such that they all randomly get lists and use eachother as slaves, but getting file distribution right in the face of a dead host is still a problem. Partial failures are a HUGE problem (if a machine blows up, how do you ask it how many people it sent to?), as is centralized logging. Even knowing when something is going wrong is hard... Mass mail is difficult to debug. Plus, I tend to have other things to do, and this works fine, now. I just want it perfect...

    As a benchmark, we send about 7 million messages a week. Delivery starts shortly after midnight; everything is done by 5AM.

    I haven't gotten in to handling bounces and bad addresses yet... let alone forged subscriptions. Suffice to say we have a complicated procmail config and a script that talks to mysql to record delivery failures, and a complicated transfer/update arrangement to update our oracle-based master list when too many delivery failures happen... and a blacklist for forgeries.

    Whew. Running mass mail servers is a mess. I don't wish that was all I had to do, but if it were, I could probably get it right.

    -Abulafia, not logged in.

  • awwww... crap. somebody musta fucked it up. I don't really want to describe what happened in detail, due to security, of course. sorry about it, tho.

    i'm not really sure what to do, but please email me at security@63.225.139.3 so that we can have this conversation in provate.
  • Postfix [postfix.org] for mail and Mailman [list.org] for the list software.
  • by Anonymous Coward
    Forgot to mention - The custom MTA _has_ to sort the addresses by domain. This is what bulk_mailer does, and it has an enourmous impact on delivery times. Also, having a caching name server with a lot of RAM dedicated to the cluster helps quite a bit. Skip it if you're low on cash, but it helps a lot. Building this is actually one of the cheapest projects I've done, cap ex wise - ~6K in hardware. Lots of engineering, though. I can't release the code, and even if I could, a lot of it is tied to sendmail/procmail/mysql config. Well, enough said. There you go. --abulafia, again.
  • I run a discussion list with about 350 subscribers (225 in 'direct mail', another 125 in digest) that gets about 100 messages from subscribers per day. It runs on sendmail & majordomo on an AMD K6-233 w/ 64MB RAM running RedHat 6.1.

    Neither myself, nor the guy who actually owns the machine, are sendmail experts, so our problems stem from sendmail taking so long to deliver messages to places having trouble recieving mail. Creates a small backlog, but messages are delivered to subscribers in under an hour usually. Subscription requests and admin things by email usually take under a minute to get confirmation.

    We also keep archives using MHonArc and glimpse, searching is a little slow, but that's to be expected considering the machine speed, and tens of thousands of messages.

    It really isn't a bad setup, considering all the software is free, the machine was sitting in the basement collecting dust, and it's run off a static IP cable modem.

  • qmail should be able to take your load.

    When I last checked, Hotmail and Yahoo mail used qmail. I doubt your load is more than theirs.

    For discussion style mail lists you can use ezmlm. But if it's just one way, one to many broadcast, just qmail will do.

    However if the problem is bandwidth, then you need to either get a larger pipe, OR find someone willing to redistribute the mail for you- e.g. you send them a single email with tons of addresses and they handle the delivery. Do NOT use someone's mail server for that without their permission.

    Good luck,

    Link.
  • You might want to have a look at PetiDomo III which is designed to handle huge mailinglists.

    http://www.cys.de/deutsch/produkte/index-petidom o.html

    The previous release PetiDomo II has been released under the GPL. Their website is in German, but if you send them an email they should be able to help you in english.

    Good luck,

    Heiko
  • by djweis ( 4792 )
    It seems to be popular with people that send hundreds of thousands of messages. I think they have a free (beer) version for a limited number, but it's very expensive as you add users.
  • Technically, it is a moderated announcement list. I send out a joke every day to every subscriber, which as I said, is quite a few. I wasn't planning on having so many - it started out to be just for family and friends, but I hope that I can quit my day job soon!

    I do need some kind of list software - I can't bcc: the joke to a thousand people using microsoft lookout or Eudora!

    Anyway, the subscribers "don't" get to post ot the list - but I've done experiments, I've heard from other people - my list package doesn't have the best security, and it might be possible for an unauthorized person to send a message to the list. Which would be bad.

    I'm running it on a 90MHz Macintosh (yes, it sucks). The package I'm using - I probably shouldn't say what it is - but it's meant for a list of about a hundred people. It sucks.

    Besides the potential security issue, it is damn slow. There are barely enough hours in a day to send out all of the messages. I need more speed.

    I do have a very fast DSL pipe. It's just a hardware and software problem. Basically, if you had a list, what s/w and h/w would you use?

    P.S. I just went to Best Buy today and bought Red Hat 6.2. Woo hoo! So, linux solutions are now acceptable. :)

    Thanks for the help!
  • For discussion style mail lists you can use ezmlm. But if it's just one way, one to many broadcast, just qmail will do.

    I definitely second the choice of qmail + ezmlm[-idx]. (-idx for the moderation features) ;)

    But for a list of any size you probably want the software to handle subscriptions and bounces automatically and securely, and ezmlm does just that by sending cryptographically secure confirmation requests and handling bounces by encoding the subscribers address in the Return-Path.

    See http://cr.yp.to/qmail.html [cr.yp.to] and http://www.qmail.org [qmail.org] for more advertising.

  • I'd stop worrying about the hardware and software to run something like this, and hire a firm to worry about it for you.

    Say for example Listbox [listbox.com] or there's always Egroups [egroups.com]. My own company's mailing list services [catalystinternet.com] aren't really ready for prime-time yet, but I do operate the announcements-only mail list for the Philadelphia Eagles, and that is about 11000 subscribers although it is one-way only.
    --

  • I ran several mailing lists for the last company I worked for. We had a 200mhz Pentium with 128 MB of RAM running RedHat 5.2. We used qmail for the delivery and Majordomo for the list management. We also ran a caching nameserver on another box to help speed things up. That machine was a Pentium 200mhz, 128 MB RAM, and RedHat 5.2.

    One of the lists was similar to yours in that it was used for broadcast and not really for discussion. We sent out a computer security newsletter once a week to about 18,000 subscribers. I think when I left the company the total had risen as high as 20,000. This same machine also handled several discussion lists. Although they were small, one of them was quite active and had about 7,500 to 8,000 subscribers.

    My experience has been that for lists this size you don't need a fast machine as much as you need RAM. Also, Sendmail doesn't seem to be as efficent as qmail. There was a noticable increase in delivery times once we moved to qmail from Sendmail. For the 18,000+ subscriber list, I kept my personal email address at the top and bottom of the mailing list addresses and there would be about a four hour delay between receiving the first message and the last one. Not too bad. I don't remember the times for Sendmail but it wasn't quite that fast.

  • I second this, I am managing an egroup mailing list, works pretty well so far, quick delivery, no spamming problems. You have even further services as Chat and archives access.
  • My recommendations follow most of the others. Get Majordomo for handling the list, and any decent MTA like sendmail or qmail for the actual sending. Make sure your machine is fairly reliable, and keep logs of everything you do. If you have problems, you probably are better off to not send out a joke one day than to send 2 or more jokes in a day trying to fix the problem.

    Since you are rightly worried about security, I'd recommend going out and reading up on everything you can. bugtraq archives, dejanews, risks archive are a few of the places to start. You've obviously put some time and effort to grow a lame joke list to 10k subscribers, so put some effort into protecting it.

    I noticed your emails coming into my /. only account starting soon after you posted this question. Did you notice someone loading up your mailing list with the complete user list scammed from /. ? Are you planning on cleaning up your list, or are you going to wait for each person to go clean it out themselves?

    the AC

Hotels are tired of getting ripped off. I checked into a hotel and they had towels from my house. -- Mark Guido

Working...