1.15. What is the largest list Mailman can run?

There are no built-in limits to the size of a Mailman list. The largest site reported to date had around 550,000 members split over four relatively low-traffic lists (If you have run a list larger than that with Mailman, please email <gward AT python DOT net> or <brad AT python DOT org> and we will update this FAQ entry.)

However, there's more to worry about than the size of your largest list; if you're concerned with system load and bandwidth consumption, then you have to consider the total population of all your lists, the activity level of each one, and the total aggregate activity. For example, 10 lists with 10,000 members each is pretty much equivalent to a single 100,000 member list - assuming the level of traffic is the same. And a 10,000-member list that gets a single announcement per week is much less load than 10 active 1,000-member discussion lists.

Some case studies:

In his initial announcement of the availability of the Mailman mailing list management program at the LISA'98 conference (see http://www.usenix.org/publications/library/proceedings/lisa98/viega.html, John Viega (the original author of Mailman), had this to say regarding performance:

 While Mailman is too new to have much hard data in the way of
 performance metrics, we do know that, given a well designed
 mailing list management system, the performance of the mail
 transport agent (MTA) will have a much more significant impact.
 We have found that even a low-end configuration can handle
 large amounts of traffic. For example, one mailing list managed
 by Mailman has had up to 3000 subscribers, and often receives
 100 messages in a day (i.e., hundreds of thousands of daily
 deliveries). The list runs on a low-end Pentium with 48MB of
 RAM. The machine runs sendmail on GNU/Linux. The machine also
 hosts an NNTP news feed for a small ISP, and is able to handle
 the load, although sendmail sometimes needs to queue messages.
 As Mailman proceeds through beta test, we plan to gather more
 detail performance data.

As of December 2001, ActiveState runs around 50 external mailing lists with a total subscriber population of about 93,500. Their largest "announce" list has just under 50,000 subscribers; they have a handful of discussion lists with around 2,000 subscribers each.

A system administrator with The Guardian (the British newspaper) described running an announce-only list with 147,000 subscribers (only used a few times a year). The same organization also has a daily distribution list with 60,000 members, and for a time had an unmoderated discussion list with 20,000 members. (No technical problem, the lawyers put an end to it.)

As of July 2004, FreeBSD.org has over 100 lists, two announce-only that have about 10,000 to 15,000 subscribers, and two discussion lists with around 5,000 subscribers (which can be pretty busy with dozens to almost a hundred messages per day), and several more busy discussion lists with around 3,000 subscribers.

As of September 2004, lists.apple.com has around 115 lists, gets 600-650 messages posted per day, resulting in around 14 million messages being delivered per month.

In February 2005, panic.com reported that they were running an announce-only list with over 120,000 recipients, and were running into some performance problems - primarily having to do with excessive contention in the Mailman/Python "pickles". However, at that time, they had yet to try splitting the mailing lists into multiple sub-lists with an umbrella list for the parent, which would allow the sub-lists to be processed in parallel. See the thread at http://mail.python.org/pipermail/mailman-users/2005-February/042506.html.

In December 2005, we received a report from a mailing list administrator running a dedicated server at HostMySite.com, associated with a well-known TV personality, having a total of about 550,000 users spread over four relatively low-traffic announcement-only lists. The administrator was looking at splitting this across 100-200 lists, to see if that would reduce Python "pickle" contention. See the thread including http://mail.python.org/pipermail/mailman-users/2005-December/048316.html for more information.

As of November 2007, python.org had 132 lists, with a total of almost 32,000 subscribers (just over 23,000 unique addresses), with a very active discussion list of more than 3500 subscribers, and a much less active announce-only list with over 3300 subscribers, and several others with a thousand or more subscribers. On a daily basis, python.org regularly receives 30,000-50,000 messages (temporarily rejecting about 30,000-50,000 incoming connections due to lack of reverse DNS, and permanently rejecting about 130,000 to 300,000 incoming connections/messages as being "spam"), and typically transmits anywhere from 225,000 to 500,000 messages. There have been more than a few days that had significantly more or less traffic (some days over a million messages per day). See also http://mail.python.org/pipermail/mailman-users/2007-September/058231.html.

SourceForge may be the largest Mailman-hosted mailing list site currently in operation, but we are still working on obtaining details of their operation.

Conclusion:

If you're going to be running lists with more than a few hundred members, you should take a look at the entries on tuning your MTA and Mailman to work best together. They're in Section 6 of this FAQ.

If you're going to be running lists with more than a few thousand members, then you need to have a thorough understanding of how your MTA and Mailman work, and a deep and intimate familiarity with Internet e-mail in general. And of course, you will have to carefully consider how best to tune your MTA and Mailman to work best together. If you're missing any of these prerequisites, you're in for a difficult time.

Note that if you're using personalization and VERP (see <4.12 What about VERP?>), this will have an impact on your system performance and will make your tuning process more difficult.

Converted from the Mailman FAQ Wizard

This is one of many Frequently Asked Questions.