[OpenAFS-devel] fileserver profiling

[chas williams - CONTRACTOR]

In message <20050314040142.GN9768@kalmia.hozed.org>,Troy Benjegerdes writes:
>I just switched to a 100mb cache, (previously it was 1GB), and ran afsd
>with "afsd -memcache -chunksize 16", and now I'm getting a better, but
>not optimal throughput of around 25MB/sec.
>
>(Out of curiosity, what cache sizes do people generally run at?)

generally speaking memcache is going to be faster.  if you have a 
good link to the fileservers (and network traffic doesnt bother 
you) then memcache is going to be the way to go.

increasing chunksize is always going to help with larger files.  if i
had to guess, i would say afs (fileserver/cache manager) was tuned to
handle a huge number of tiny files.  since performance on tiny files
is always dominated by system overhead, the fileserver's performance
'problems' aren't a big deal.  however, when it comes to large files
you start to see trouble.

from results from bonnie i made way back:

Version 1.02c       ------Sequential Output------ --Sequential Input- --Random-
		    -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
relax/memcache=18m, chunksize=13 (default)
	       256M  3669  68  7695  33  1413   7  1566  24  1739   2 218.5   2
relax/memcache=18m,chunksize=14
	       256M  3870  68  8507  32  2101   9  2663  40  2912   4  77.0   1
relax/memcache=18m,chunksize=15
	       256M  3968  70  8966  32  2935  12  3381  50  4310   5  96.7   1