72497 - evaluate lea malloc-2.7.0

Status: RESOLVED WONTFIX

(Core :: XPCOM, defect, P2)

Description

[Chris Waterson]

From private email:

Hi; If you are still using prereleass, you might want to
grab the 2.7.0 release:
  http://gee.cs.oswego.edu/pub/misc/malloc.c

I suspect that it is marginally faster in mozilla than pre6.

Also, while I couldn't replicate your footprint tests (well, not
without more time to collect and cache pages, etc needed to do it
right), I did recheck the trace you put up. A plot of it (among others)
is at http://gee.cs.oswego.edu/dl/malloc-plots/index.html

-- Doug Lea

Comment 1

[Chris Waterson]

Nominating for mozilla0.9. edward, could you please try to look at this? Also,
Doug may have ``fixed'' bug 73053 without forcing us to apply our own mods.

Comment 2

[Randell Jesup [:jesup] (needinfo me)]

Bug 73053 is not covered by this bug or Doug's updates to 2.7.0 release.  I
suspect he's only tested the pthread code under Linux.

Comment 3

[Randell Jesup [:jesup] (needinfo me)]

Reassigning to waterson (Chris, feel free to reassign it again).

Comment 4

[Chris Waterson]

mcafee, could you take a look at this?

Comment 5

[Chris McAfee]

-> dp

Comment 6

[Adam D. Moss]

The ptmalloc page <http://www.malloc.de/en/index.html> seems
to imply that an older-but-more-thread-friendly adaptation of
lea-malloc is used in glibc.  Performancewise it could be
a toss-up between the improvements added to lea-malloc since
the fork and the overheads of its very coarse locking in
comparison to ptmalloc/glibc.  I'm certainly very interested
in seeing the results!

Comment 7

[Chris Waterson]

shaver's been doing some work on this, I think.

Comment 8

[Suresh Duddi (gone)]

Shaver wanna take this over ?

Comment 9

[Mike Shaver (:shaver -- probably not reading bugmail closely)]

Yeah, I got numbers at one point but they were pretty noisy.  I'll run again soon.

Comment 10

[Randell Jesup [:jesup] (needinfo me)]

Untargeting from 0.9.8

Comment 11

[Mike Shaver (:shaver -- probably not reading bugmail closely)]

glibc2.3 is taking the new ptmalloc.  We should probably look at taking it too.

1.0, I say.

Comment 12

[Mike Shaver (:shaver -- probably not reading bugmail closely)]

We're not going to take a new allocator for 1.0, and newer glibcs will soon have
the sexy new ptmalloc anyway.

We'll revisit this with dp when he gets back, or something.