running ddsnap on 2.6.24.2 kernel seems slow

[Jiaying Zhang]

I found the performance of ddsnap on our latest 2.6.24.2 kernel
is noticeable slower than that on the 2.6.22.18 kernel. For example,
with the 2.6.22.18 kernel, nsnaps takes about 35s on a Dell2950
and 150s on a HP xw4100. But with the 2.6.24.2 kernel, nsnaps
takes about 50s on a Dell2950 and 180s on a HPxw4100.

We made some changes on the bio_throttling patch in the 2.6.24.2 kernel.
I tried the same test without that patch but did not see any performance
improvements. The two kernel configurations also look similar. I don't
know if the problem is caused by the bio changes added in 2.6.24.2
mainline. Does anyone experience the similar problem?

Jiaying

[Will Nowak]

Can we be sure we are not having negative issues as a result of the
CFS introduction in 2.6.23? There are a couple Ubuntu bugs floating
around based on a poorly decided kernel configuration.
http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=blob;f=Documentation/sched-design-CFS.txt;h=88bcb87673354302737be5bbaac5b8c049941d86;hb=HEAD

-Will

[Jeff Schroeder]

On Tue, May 20, 2008 at 3:36 PM, Jiaying Zhang <jiay...@google.com> wrote:

jeff@jschroeder2:~/src/software/zumastor/kernel/config$ grep
FAIR_GROUP_SCHED 2.6.22.18-i386-full 2.6.24.2-amd64-full
2.6.24.2-amd64-full:CONFIG_FAIR_GROUP_SCHED=y

You might try disabling CONFIG_FAIR_GROUP_SCHED in the kernel config,
rebuild it, and test again. That has a lot of known
regressions. Does that speed it up any?

>
> Jiaying

--
Jeff Schroeder

Don't drink and derive, alcohol and analysis don't mix.
http://www.digitalprognosis.com

[Jiaying Zhang]

I tried the 2.6.24.2 kernel with CONFIG_FAIR_GROUP_SCHED disabled, but
still did not see any improvements.

Jiaying

[Will Nowak]

You might also try disabling CONFIG_FAIR_USER_SCHED
-Will

[Jiaying Zhang]

CONFIG_FAIR_USER_SCHED was automatically turned off with
CONFIG_FAIR_GROUP_SCHED disabled.

Jiaying

[Will Nowak]

ah, ok then. I'm out of ideas :)
-Will