Charles Wright wrote: > Here's an update: > > I thought that the error message > arcmsr0: too many outstanding commands > might be due to a Scsi queue being over ran >
Rather than messing with sd_max_throttle, you might try changing the number of iops ZFS will queue to a vdev. IMHO this is easier to correlate because the ZFS tunable is closer to the application than an sd tunable. Details at: http://www.solarisinternals.com/wiki/index.php/ZFS_Evil_Tuning_Guide#Device_I.2FO_Queue_Size_.28I.2FO_Concurrency.29 iostat will show the number of iops queued to the device in the actv column, but for modern hardware this number can fluctuate quite a bit in a 1-second sample period -- which implies that you need lots of load to see it. The problem is that if lots of load makes it fall over, then the load will be automatically reduced -- causing you to chase your tail. It should be fairly easy to whip up a dtrace script which would quantize the queue depth, though... [need more days in the hour...] -- richard _______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
