raid + xfs
hi. i'm using xfs on top of raid, and have noticed some unusual behavior. my basic hardware configuration is 850 p3, intel d815eea2 motherboard, 4 ibm drives, 2 promise cards, and a separate drive on built in ide channel as the root (this was not the initial configuration, but kind of evolved as we went ahead.) what we noticed was that the resyncing speed as reported by cat /proc/mdstat was originally 100k/sec, making resyncing take forever. we bumped /proc/sys/dev/raid up, and got resyncing to around 3000k/sec, with 90% cpu usage for raid5sync. then i decided to run the bonnie benchmark, and *while* bonnie was running, resyncing went up to 1k, with low CPU usage. when bonnie finised, resyncing started to behave badly again. (and bonnie's performance wasn't noticably impacted by the resyncing either). perhaps in a few weeks i might be able to investigate the raid code, but i'm hoping that someone might know what is causing this behavior. please CC me at [EMAIL PROTECTED] if you reply. thanks, jon bright -- jonathan bright [EMAIL PROTECTED] 415.820.7322 - voicemail/fax www.brightconsulting.com Binary data Binary data Binary data
raid + xfs
hi. i'm using xfs on top of raid, and have noticed some unusual behavior. my basic hardware configuration is 850 p3, intel d815eea2 motherboard, 4 ibm drives, 2 promise cards, and a separate drive on built in ide channel as the root (this was not the initial configuration, but kind of evolved as we went ahead.) what we noticed was that the resyncing speed as reported by cat /proc/mdstat was originally 100k/sec, making resyncing take forever. we bumped /proc/sys/dev/raid up, and got resyncing to around 3000k/sec, with 90% cpu usage for raid5sync. then i decided to run the bonnie benchmark, and *while* bonnie was running, resyncing went up to 1k, with low CPU usage. when bonnie finised, resyncing started to behave badly again. (and bonnie's performance wasn't noticably impacted by the resyncing either). perhaps in a few weeks i might be able to investigate the raid code, but i'm hoping that someone might know what is causing this behavior. please CC me at [EMAIL PROTECTED] if you reply. thanks, jon bright -- jonathan bright [EMAIL PROTECTED] 415.820.7322 - voicemail/fax www.brightconsulting.com Binary data Binary data Binary data