[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: >1024 connections in slapd / select->poll

--On Wednesday, November 17, 2004 10:34 PM -0500 Wesley D Craig <wes@umich.edu> wrote:

In fact, we've currently got the MTA (it's called simta) set to limit
itself to 250 inbound SMTP connections.  We monitor the SMTP interface
(and the LDAP interface) to ensure that it remains responsive.  Our
customer expectation is that when email is sent, it will arrive at the
UMich recipients mailbox within seconds.  In short, the concurrency
reflects how our users send & receive email.

Certainly an interesting way to do it. Stanford processes around the same amount of email each day, and we definitely do not do it this way.

Basically, we have 6 "incoming" email routers, that are used to deliver incoming email. We also have 4 outgoing smtp servers. If an email sent to an SMTP server is for an @stanford.edu address, it also handles the delivery, rather than passing it on to the incoming mail routers.

We use sendmail (with custom patches to handle GSSAPI binds to the LDAP servers) as our MTA. We've never hit any problem with 1000+ open connections. We do dedicate 3 servers solely to mail handling, but they are all in a load-balance pool, and on a few occasions, I have dropped 2 of the 3 servers from the pool, and the remaining server handled the load just fine.

The following link shows a graph of binds/second tracked in cricket for the 3 servers:


As you can see, there is rarely more than 40 active bonds on the most heavily loaded server in any given second, and if you added them all together, you would still only be averaging 90-100 binds/second. All the lookups tend to take less than 2 seconds.

So I'd say that keeping 250 connections open per server is extreme overkill.


Quanah Gibson-Mount
Principal Software Developer
ITSS/Shared Services
Stanford University
GnuPG Public Key: http://www.stanford.edu/~quanah/pgp.html