OSDL tiobench Sequential Reads improved with AS due to read-ahead changes
Hello;
I have compiled the results for tiobench with Anticipatory Scheduler for before and after the addition of the readahead-revert-lazy-readahead patch
in the 2.6.1-mm4 kernel. I have found our SCSI systems showed up to 50%
improvement in Sequential Reads on ext2. (2-CPU/MegaRAID/RAID0/5disks and 4-CPU/AACRAID/RAID0/5disks). Here are the plots:
http://developer.osdl.org/judith/tiobench/big_jump_SR_24CPU/sr.html
Oddly, random reads also show slight improvement:
http://developer.osdl.org/judith/tiobench/big_jump_SR_24CPU/rr.html
Our 2CPU/MegaRAID particularly has been underperforming in the past, but now is much more comparable to 'deadline'. Here are plots of the current -mm and mailine kernels:
http://developer.osdl.org/judith/tiobench/2CPU_big_jump_SR/sr.html
Thanks;
Judith Lebzelter
OSDL