[Prev][Next][Index][Thread]

Re: E/L 1.99n broken under kernel 1.2.10



>>>>> "Tom" == Tom Bartol <bartol@salk.edu> writes:

    Tom> Hi all, I want to second the observation that E/L 1.99n seems
    Tom> to have a memory leak under kernel 1.2.10.

Right, that's what Art was running.  Mat and I are running 1.2.0 and
I'm not sure what Cotton is running.

    Tom> I haven't
    Tom> registered for executor yet (plan to soon, I was waiting to
    Tom> see what 1.99n was like first) so I ftp'd 1.99n from
    Tom> vorlon.mit.edu, installed it (BTW I like the makefile
    Tom> installation procedure), and fired it up.  I get the splash
    Tom> screen and hit "demo" and then watched my xsysinfo swap usage
    Tom> indicator grow and grow and GROW.

I just ran E/L 1.99n over here while running top.  I think we have a
small memory leak each time a new application starts, but it's on the
order of 1k per application -- any leak is a bad leak, but we're
definitely seeing different memory footprints on our different
kernels.

    Tom> Finally I just killed the
    Tom> process.  Thinking that perhaps incompatible bits of 1.99m
    Tom> were left hanging around after installing 1.99n over the top
    Tom> of it, rm'd the entire executor install tree and installed
    Tom> 1.99n clean and fresh with the same final results.  So, like
    Tom> I said I am planning to register right away and I would be
    Tom> more than happy help ARDI focus on squashing this bug.  I
    Tom> imagine it really is something having to do with kernel
    Tom> 1.2.10.  Tonight I'll switch back to 1.2.8 and also to 1.2.11
    Tom> (or 1.2.12 if such an update comes out later today) to try to
    Tom> narrow this down a bit.

That would be very handy.  If anyone out there is running 1.2.10 and
*not* seeing the memory leak, that would be handy as well as any
information from anyone running > 1.2.10.

    Tom> I running on a 486/66 with 16 Megs,
    Tom> ATI Mach32 Graphics UltraPro with 2 Megs, and Slackware
    Tom> 2.3.0.

Mat and I both use ATI Mach32 cards on 90 MHz P5s.  Mat's machine has
32 MB, mine, 24.  I think he has 2 MB VRAM, and I have 1 MB.

    Tom> And while I'm at it, I'd like to take this
    Tom> opportunity to offer my most profuse thanks and praise to
    Tom> ARDI for such a killer product! (my probelms with 1.99n
    Tom> notwithstanding).

Er, I'm not sure that "killer" is the best adjective for a program
that apparently consumes vast amounts of memory under some kernels.
Too early to tell whether the kernel hypothesis is true, but if it is,
and it's our fault, you can see how this could have slipped by.

	--Cliff
	ctm@ardi.com


Follow-Ups: References: