New malloc ready, take 42
phk at phk.freebsd.dk
Fri Dec 23 01:42:17 PST 2005
In message <E1E22E2D-A6D5-49CC-9649-C37C50F0443B at freebsd.org>, Jason Evans writes:
>So, how about it? Is jemalloc ready to go in now?
Sounds like it to me :-)
Various fatherly advice:
During the first part of the integration period performance is not
important, correctness is.
Enable all reasonable checks. If you have something like phkmallocs
EXTRA_SANITY, turn it on.
There will be whineage about performance and how FreeBSD is dying etc,
just ignore it, those people havn't got a clue and you'll get to show
them later when the checks are turned off.
I also advice that you to keep both mallocs in the tree for a
transition period and define an environment variable that causes
fallback to phkmalloc. That way you don't risk stopping people
dead in their tracks if some unforseen sideeffect shows up.
Regarding unforseen sideeffects, the thing I would worry most about
is alignment, where phkmalloc is very generous.
And once again: Thankyou!
Poul-Henning Kamp | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG | TCP/IP since RFC 956
FreeBSD committer | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.
More information about the freebsd-current