tomcat-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Someone Else" <>
Subject Re: Significant Deployment Issue - Unsolvable?
Date Tue, 22 Jun 2004 20:51:12 GMT
Hi QM,

Thanks for your suggestions.

Our system is running 150 similar web apps - they are sharing the codebase - 
only the jsps are different.

We've done significant profiling and discovered only the following:

There are a lot of byte arrays that are being produced. We looked into this 
as it could be a mem leak, but no specific operation created a whole bunch 
of them at one time. They were being created in some sort of input and 
output buffer classes - I have no idea why. We rely exclusively on GC to 
clean up used objects, but as I understand it the servlet end of life will 
free everything for GC so this doesn't appear to be the issue.

Either way, on our simulated system the mem usage stabilized and did not 
grow after a certain point and we could not get it to crash.

Our test system only has 2 virtual hosts instead of 150. Also, it runs in a 
much more restricted mem environment and stabilizes around 30 MB usage.

Splitting the instances would work well if there were 150 different web 
apps, but since they are all essentially the same, multiple 
instances/machines would only prolong the expactancy, but would not solve 
the problem.

Any suggestions?


>From: QM <>
>Reply-To: "Tomcat Users List" <>
>To: Tomcat Users List <>
>Subject: Re: Significant Deployment Issue - Unsolvable?
>Date: Tue, 22 Jun 2004 15:18:41 -0500
>On Tue, Jun 22, 2004 at 12:03:17PM -0600, Someone Else wrote:
>: The memory use grows perdictably until it reaches around 20 hours, then 
>: starts to gobble up RAM until it crashes, which I don't understand. The
>: stangest thing is that I've allocated 1GB mem for the Tomcat instance, 
>: it only uses up to ~400MB total heap (used+free) before crashing.
>You've already ruled out DB connections and threads.
>All that's really left to do is run a profiler: that will show what's
>going on, where.
>Would it be at all possible to split your deployment into, say, 2 groups
>of 50 vhosts or 3 groups of 50?  That would help whittle down whether
>there's a problem in a specific webapp that's bringing down the whole
>lot of them.
>software  --
>tech news --
>To unsubscribe, e-mail:
>For additional commands, e-mail:

MSN Premium: Up to 11 personalized e-mail addresses and 2 months FREE*

To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message