Re: [PATCH v2] proc: speedup /proc/stat handling

[KAMEZAWA Hiroyuki]

On Wed, 25 Jan 2012 17:04:16 -0800
Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:

> On Wed, 25 Jan 2012 06:29:32 +0100
> Eric Dumazet <eric.dumazet@xxxxxxxxx> wrote:
> 
> > Le mardi 24 janvier 2012 __ 17:27 -0800, Andrew Morton a __crit :
> > 
> > > I had a fiddle on an 8-way x86_64 machine.  I'm unable to demonstrate
> > > any improvement for either of
> > > 
> > > time (for i in $(seq 1000); do; cat /proc/self/stat > /dev/null; done)
> > > time (for i in $(seq 1000); do; cat /proc/1/stat > /dev/null; done)
> > > 
> > > oh well.
> > 
> > What size is /proc/stat ?
> 
> About 40mm, but it depends on the font size.
> 
> > wc -c /proc/stat
> > 
> > If under 4096, there is no problem with existing code.
> 
> akpm2:/home/akpm> wc -c /proc/stat 
> 2800 /proc/stat
> 
> > I had the problem on a 16-way machine.
> 
> OK..


I wrote following patch just for my fun, which makes /proc/stat twice fast.
But I'm not sure whether this kind of dirty && special printk is worth to do or not..
because I can't see /proc/stat cost at shell-scripting.

With python script test (read /proc/stat 1000times)
==
#!/usr/bin/python

num = 0

with open("/proc/stat") as f:
        while num < 1000 :
                data = f.read()
                f.seek(0, 0)
                num = num + 1
==

*Before patch
[kamezawa@bluextal test]$ time ./stat_check.py

real    0m0.154s
user    0m0.020s
sys     0m0.130s

*After patch
[kamezawa@bluextal test]$ time ./stat_check.py

real    0m0.080s
user    0m0.029s
sys     0m0.048s


For ascii interface, format_decode() and number() seems to be
very costly...

==