On Thu, 28 Sep 2000, Hauke Fath wrote:

> Hi,
> 
> while I am listening to my workstation swapping itself to death...
> 
> I've got IBM's Java compiler Jikes 1.0.6 set up on a 1.4.1 server 
> (PPro 200, 128 MB RAM) to do regular test builds from the current 
> repository. During compiles, the load goes up to 3.5, and Jikes eats 
> up to 50MB memory. The whole build (cvs checkout, compile, jar) takes 
> about five minutes. If anybody is doing anything on the machine at 
> the time (it's an SMB/AppleShare server, CVS server, html documents 
> server), they feel the load but can still work.
> 
> Now, I have set up Jikes on my 1.5 workstation (2x PII/350, 256MB 
> RAM). Built from the same sources, the 1.5 ELF Jikes wants ~300MB 
> memory. The whole build takes 10..15 minutes, most of which appears 
> to be excessive swapping (cpu activity is near zero). During the 
> build, the machine is close to unusable.
> 
> Anybody seen anything like that? Any ideas?

	Can you test the 1.4.1 binary on the 1.5 system to confirm it
	is a library/toolchain rather than a kernel issue (copying
	a.out libraries across into /usr/lib/aout and using the
	uuencode a.out ld.so from 1.5).

	Assuming it is userland/toolchain, could it be the new malloc
	library - does "ln -sf R /etc/malloc.conf" have any effect?

                David/absolute
			       -- www.netbsd.org: A pmap for every occasion --