tomcat-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Christopher Schultz <>
Subject Re: Tomcat with 8 GB memory
Date Thu, 26 Jul 2007 20:36:30 GMT
Hash: SHA1


lightbulb432 wrote:
> When I asked this question I had in mind a few potential downsides to one 8
> GB Tomcat. One is memory leaks: if Tomcat starts to leak memory (not sure if
> this happens too often, if ever, but I'm speaking purely hypothetically)
> then you'd rather not want your single Tomcat to eat up all the memory. If
> you had two Tomcat instances running the exact same application, then the
> memory leak is only affecting up to 4 GB?

Well, if you have them properly load-balanced, then they'll both run out
of memory at roughly the same time, right?

> The other is high-availability - if you have a huge 8 GB box and it's
> because you decide not to have 4 x 2 GB smaller boxes, then you definitely
> cannot get high-availability against hardware issues - but say you're
> pushing out an upgrade to the application (i.e. the WAR file). If you have
> two Tomcats, couldn't you bring one down while the other continues to serve
> requests, push out the upgrade, bring it back up, then do the same with the
> other application? With only one instance, you must have downtime right?

True. You only asked about performance. In terms of performance, fewer
moving parts is always better.

Distribution helps solve other problems like geographical load-balancing
to avoid network bottlenecks, improve availability (the more servers you
have, the more likely one of them is to be in service), and facilitate
roll-outs (upgrade one at a time). I generally find that the database
upgrade is the thing that kills you, and that you need to have at least
/some/ total down time to get everything in sync... re-deploying an
application (upgrading) takes much less time than changes to your database).

> The last is at the following link, several performance graphs for Linux show
> a slight or sharp decrease in performance for a given Tomcat instance beyond
> a certain number of users! Why is this? Wouldn't these diminishing returns,
> if true, severely cripple one 8 GB instance serving 2*n users compared to two
> 4 GB instances serving only n users?

That's a good question. Having only looked at the graph and not read the
article in detail (or at all, really), I can only imagine that Linux and
Windows have different strategies of dealing with incoming connections
that the server software refuses to accept them. Perhaps Linux maintains
a larger queue of incoming connections than Windows does, resulting in
more successful connections, but slower response time for them.

- -chris

Version: GnuPG v1.4.7 (MingW32)
Comment: Using GnuPG with Mozilla -


To start a new topic, e-mail:
To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message