Re: performance degraded 2.4.8 -> 2.4.9?

[Dan Yocum <yocum@xxxxxxxx>]

Dan Yocum wrote:

> I'm doing my bonnie++ tests on 2.4.15pre4 at the moment and I'm seeing a
> little better performance than 2.4.9 (for writes), but it's still not at the
> 2.4.5 level.  I'll let it run a few more iterations before I say what the
> numbers are.


Bonnie++ on 2.4.15pre4:

Version  1.01b      ------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
dp15         2G:64k           19484  10 14107  12           37131  15
234.5   7
dp15         2G:64k           23628  13 19335  16           66596  32
300.0   8
dp15         2G:64k           20503  11 21895  18           65978  32
302.2   9
dp15         2G:64k           23464  13 19560  16           69591  33
297.9   8
dp15         2G:64k           20970  12 21602  18           66847  32
309.2   7
....


dd on 2.4.15pre4:

[root@sdssdp15 data]# time dd if=/dev/zero of=/dev/md0 bs=1024 count=2000000
2000000+0 records in
2000000+0 records out

real    1m55.272s
user    0m1.210s
sys     0m22.810s

Or, 17.8MB/s 

:-/



Just for grins, if anyone is interested:

Bonnie++ on 2.4.5 compiled with kgcc:

Version  1.01b      ------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
dp15         2G:64k           33613  20 14157  14           93517  47 221.6 
7
dp15         2G:64k           29749  17 14715  14           98281  48 239.4 
7
dp15         2G:64k           32543  19 15088  14           116945  55 212.0
6 dp15         2G:64k           30620  19 15411  14           110586  53
216.1 7 dp15         2G:64k           31232  19 14447  14           119500 
59 204.5 6 dp15         2G:64k           29875  18 15458  15          
113109  62 212.1 7
...

Bonnie++ on 2.4.5 compiled with gcc-2.96-85:

Version  1.01b      ------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
dp15         2G:64k           34800  18 17905  15           145748  60 218.5
5 dp15         2G:64k           34439  18 17338  15           167293  69
216.2 5 dp15         2G:64k           34739  18 17621  15           161592 
65 215.4 5 dp15         2G:64k           34831  18 17406  15          
163942  64 208.5 7 dp15         2G:64k           34883  18 17105 
14           151747  67 213.9 6 dp15         2G:64k           33596  18
17657  14           145274  63 208.1 6
...

Not only does the gcc compiled kernel give faster I/O, the results vary
less, too.

Cheers,
Dan


-- 
Dan Yocum
Sloan Digital Sky Survey, Fermilab  630.840.6509
yocum@xxxxxxxx, http://www.sdss.org
SDSS.  Mapping the Universe.