[Planetlab-users] RE: java memory overconsumption
Marc E. Fiuczynski
mef at CS.Princeton.EDU
Wed Aug 4 16:07:41 EDT 2004
Yes, it is your idsl_pepr slice that appears to be overconsuming memory
From: users-bounces at planet-lab.org
[mailto:users-bounces at planet-lab.org]On Behalf Of BRETT, PAUL
Sent: Wednesday, August 04, 2004 1:56 PM
To: Marc E. Fiuczynski; Adams, Robert; Bowman, Mic; Rob Knauerhase
Cc: users at lists.planet-lab.org
Subject: [Planetlab-users] RE: java memory overconsumption
Which slice is causing you concern, and what is PlanetLab policy on
For example, currently the idsl_pepr slice runs on about 40 nodes at any
time, with a memory consumption of 100MB. It would be fairly simple to
refactor this as 80 nodes, each with a 50MB footprint, which might take
us off the top 10 list. Would this be a more acceptable solution :-)
The main application we are running in the PEPR slice is an off the
shelf jabber server implementation, written in Java. We are taking a
look at the memory footprint of the application, but in the mean time
I've cc'ed the users list on this email to see if we can share 'best
practices' on optimizing the JVM for reduced memory usage. Maybe we
need a Wiki page?
Email: paul.brett at intel.com
Tel No: +1 503 712 4520
From: Marc E. Fiuczynski [mailto:mef at CS.Princeton.EDU]
Sent: Wednesday, August 04, 2004 8:49 AM
To: Adams, Robert; Bowman, Mic; BRETT, PAUL; Rob Knauerhase
Subject: java memory overconsumption
Our nodes are in trouble, because they are running out of swap space.
slice is listed (on comon) as one of the top memory consumers across
nodes (using significantly more than 100MB in many cases).
>From what I remember, when trumpet was written in perl it wasn't such a
memory hog. But it appears that you have re-written it into java and the
footprint of your apps has swelled. In general, it appears that many
researchers are using java and started pushing those apps during this
PlanetLab simply does not have the resources to support such large apps.
Please do whatever you can to minimize your memory footprint.
Users mailing list: Users at lists.planet-lab.org
More information about the Users