re: WABPL tstile/performance issues...[LONG]

[matthew green <mrg%eterna.com.au@localhost>]

   On Sat, Jan 31, 2009 at 05:45:21PM -0600, Greg Oster wrote:
   
   > Additional thoughts?  Is nuking the FSYNC_RECLAIM the right check?
   > (the above code will still do the right thing if FSYNC_RECLAIM is 
   > set, since it will still be keying off of anything being in the 
   > &vp->v_dirtyblkhd list)
   
   I can't see any good reason for the FSYNC_RECLAIM check.

for my simple test case, "make depend" on a dual processor ultra60,
i see a significant improvement with this patch applied:


-current:

175.148u 73.308s 2:22.59 174.2% 0+0k 0+710io 999pf+0w
176.455u 71.882s 2:24.27 172.1% 0+0k 0+761io 999pf+0w
175.614u 72.534s 2:23.06 173.4% 0+0k 0+746io 999pf+0w

-current + patch:

176.574u 70.956s 2:06.02 196.4% 0+0k 0+62io 999pf+0w
177.403u 69.977s 2:05.74 196.7% 0+0k 0+63io 999pf+0w
178.439u 72.903s 2:07.56 197.0% 0+0k 0+63io 999pf+0w



.mrg.