success with atomic kmap patches

success with atomic kmap patches

[Dave Hansen]

I was using a little script to warm up my file cache for Specweb runs 
when Martin Bligh said that it might be a good test of your kmap 
patches.  I ran 8 greps in parallel (1-per-cpu) through a 10-gigabyte 
Specweb file set which is on a RAID array.  The RAID is how I can do a 
cold run in 3 minutes 30 sec :).  Each grep works on a disjoint set of 
data.

Here's a run with the cache already warm:
http://www.sr71.net/~specweb99/run-pgrepwarm-2.5.26+lm+kmapfun-07-18-2002-18.24.38/
You'll probably only care about greptime.total, and lockstat.  The 
network stuff is cruft from when I actually run Specweb.

Here's a cold cache run:
http://www.sr71.net/~specweb99/run-pgrep-cold-2.5.26+lm+kmapfun-07-18-2002-18.25.55/

Here are warm and cold, without the kmap patches
http://www.sr71.net/~specweb99/run-pgrep-cold-2.5.26+lm-07-18-2002-18.46.27/
http://www.sr71.net/~specweb99/run-pgrep-warm-2.5.26+lm-07-18-2002-18.56.09/

I would give you oprofile data, but it appears that NMIs wreak havoc 
on a certain vendor's hardware.  When oprofile is compiled in, my box 
gets quite unstable.
-- 
Dave Hansen
haveblue@us.ibm.com