Ingo Molnar wrote:
* Jorge Nerin <jnerin@svalero.es> wrote:
Hello, I was testing the value of "2" in /proc/sys/net/tux/compression
when I found that:
Feb 24 15:17:56 head kernel: deflate in loop returned -5
Feb 24 15:17:56 head kernel: kernel BUG at gzip.c:3334!
i've uploaded a new Tux patch to:
redhat.com/~mingo/TUX-patches/tux3-2.6.3-B2
i've fixed all bugs in the gzip code i could find: made it use the 2.6
kernel's deflate library, and fixed a couple of bad assumptions. Memory
consumption should be lower as well, since now deflate state is
per-Tux-thread, not per-request. I've added the gzip fix from Miles Elam
as well.
i've done light testing with compression=2, but YMMV - does it work for
you?
i suspect the compression level should be configurable as well. Right
now it's hardcoded to level 6 - what level would be the best?
(this patch includes some more fixes as well - eg. stack footprint
reduction for CGIs.).
Ingo
I still can't upgrade to 2.6, this box has a beta redhat 6.9.5 heavily
tweaked and I can't seem to find the time to upgrade to 2.6 (the glibc
threads upgrade scares me), if you can make a patch to 2.4 I will try it.
About the compression level, the easy answer is to make it a
configurable value, either at compile time, or better like a proc entry,
but I find that 6 is a good answer, althougth the cvs folks always
recomends 3.
--
Jorge Nerin
<coma@redvip.homelinux.net>