How's linux-ibm-jdk14 in the memory footprint department?

[Nicolas Gieczewski]

Many thanks to Thomas and Dan for their replies. The actual RSS has always been around 16 MB, but I was thinking the difference between VSZ and RES was going into the swap area. Obviously a lack of knowledge on memory management on my part. I'll have to read some stuff on this.

Thanks again,

Nick


----- Original Message ----- 
From: "Sparrevohn, Thomas" <thomas.sparrevohn at eds.com>
To: "'Dan Nelson'" <dnelson at allantgroup.com>; <freebsd-java at freebsd.org>
Sent: Friday, December 12, 2003 08:02
Subject: RE: How's linux-ibm-jdk14 in the memory footprint department?


Yes - I looked at It and it seems that while it uses a large address space the RSS stays around 21MB-41MB for j2ee as an example. There are a potential problem with the way it is done that would occur if a very large library was used. Has anybody looked at deferred loads?


-----Original Message-----
From: owner-freebsd-java at freebsd.org [mailto:owner-freebsd-java at freebsd.org] On Behalf Of Dan Nelson
Sent: 11 December 2003 23:41
To: freebsd-java at freebsd.org
Subject: Re: How's linux-ibm-jdk14 in the memory footprint department?


Nicolas Gieczewski wrote:
> Thanks for the suggestion. I could only get it down to about 152 MB
> with an initial heap size of 8 MB and a maximum size of 16 MB. I 
> believe this is probably too small, though (actually, I have no idea, 
> guess I'll have to see how often the GC is kicking in). Even though 
> not all of the 152 MB (or 203 MB, for that matter) are in main memory 
> at any given time, don't they become unavailable to the OS, anyway? 

Definitely not.  Note that java will mmap any .jar files you use, so 
things like rt.jar will add 30MB to your process size right off the bat. 
    Very little of it will actually get paged in from disk, though, and 
what little does get paged in will be shared across all java processes. 
  I don't know where the rest of the 100 or so MB is going, but the most 
important thing to watch is the RSS value and whether you are swapping 
or not.  I did a simple test of running /usr/local/jdk1.4.2/bin/jar on a 
little file and pausing it so I could see the stats in top:

   PID USERNAME  PRI NICE   SIZE    RES STATE TIME   WCPU    CPU COMMAND
24511 dan        97    0   198M  7460K STOP  0:00  0.00%  0.00% jar

Note that the process space is almost 200MB, but really under 8MB is 
being used.  My guess is the extra space has something to do with 
threads, or maybe a sparse array was allocated (like what rpc.statd does).


-- 
Dan Nelson
dnelson at allantgroup.com