On Jan 9, 2008, at 9:09 PM, Rob Logan wrote: > > fun example that shows NCQ lowers wait and %w, but doesn't have > much impact on final speed. [scrubbing, devs reordered for clarity]
Here are the results i found when comparing random reads vs. sequential reads for NCQ: http://blogs.sun.com/erickustarz/entry/ncq_performance_analysis eric > > extended device statistics > device r/s w/s kr/s kw/s wait actv svc_t %w %b > sd2 454.7 0.0 47168.0 0.0 0.0 5.7 12.6 0 74 > sd4 440.7 0.0 45825.9 0.0 0.0 5.5 12.4 0 78 > sd6 445.7 0.0 46239.2 0.0 0.0 6.6 14.7 0 79 > sd7 452.7 0.0 46850.7 0.0 0.0 6.0 13.3 0 79 > sd8 460.7 0.0 46947.7 0.0 0.0 5.5 11.8 0 73 > sd3 426.7 0.0 43726.4 0.0 5.6 0.8 14.9 73 79 > sd5 424.7 0.0 44456.4 0.0 6.6 0.9 17.7 83 90 > sd9 430.7 0.0 44266.5 0.0 5.8 0.8 15.5 78 84 > sd10 421.7 0.0 44451.4 0.0 6.3 0.9 17.1 80 87 > sd11 421.7 0.0 44196.1 0.0 5.8 0.8 15.8 75 80 > > capacity operations bandwidth > pool used avail read write read write > ---------- ----- ----- ----- ----- ----- ----- > z 1.06T 3.81T 2.92K 0 360M 0 > raidz1 564G 2.86T 1.51K 0 187M 0 > c0t1d0 - - 457 0 47.3M 0 > c1t1d0 - - 457 0 47.4M 0 > c0t6d0 - - 456 0 47.4M 0 > c0t4d0 - - 458 0 47.4M 0 > c1t3d0 - - 463 0 47.3M 0 > raidz1 518G 970G 1.40K 0 174M 0 > c1t4d0 - - 434 0 44.7M 0 > c1t6d0 - - 433 0 45.3M 0 > c0t3d0 - - 445 0 45.3M 0 > c1t5d0 - - 427 0 44.4M 0 > c0t5d0 - - 424 0 44.3M 0 > ---------- ----- ----- ----- ----- ----- ----- > > > > > _______________________________________________ > zfs-discuss mailing list > zfs-discuss@opensolaris.org > http://mail.opensolaris.org/mailman/listinfo/zfs-discuss _______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
