Hi Phil, Phillip Bennett wrote: > I see in the archives there are questions and statements regarding the > partition table being mis-aligned when using AoE/vblade and this causing > more reads than needed. I used iostat to monitor at a disk I was writing to > over the network (dedicated switch, balance-rr bonding) and saw that there > seemed to be about 20+ times more READS than there was WRITES! I believe > this is the problem people were talking about, so is there a fix, or some > special way to get around this? Currently I can get 1.4Gb/sec over the > network (iperf) andat least 110+MB/sec with hdparm on the AoE device. > However, if I try to copy anything over to the disk, it can hit a maximum of > about 20-27MB/sec. Is this normal?
This sounds like you're hitting the cache alignment problem. The ridiculous amount of reads is a cache backfill penalty. I did a technical write-up of the problem, which is available at: http://nemesis.damnit.us/~khudson/text/aoe-caching-alignment.pdf let me know if you have any questions. Alternatively to changing your disk's geometry, you can turn off caching on the target by opening the exported block devices with O_DIRECT (however this has its own performance implications). Not sure how you accomplish this with vblade or if it's even possible without a source code hack. I've long since switched to qaoed. -kelsey ------------------------------------------------------------------------- Sponsored by: SourceForge.net Community Choice Awards: VOTE NOW! Studies have shown that voting for your favorite open source project, along with a healthy diet, reduces your potential for chronic lameness and boredom. Vote Now at http://www.sourceforge.net/community/cca08 _______________________________________________ Aoetools-discuss mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/aoetools-discuss
