There is a variable called zfs_max_vdev_pending. It sets the queue size for all disks. I only know the solaris well, but I know there is something similar in linux. It is set to 10 by default. You may want to up this to 2-5 per disk queue. With the 10, and 8 disks, it is pretty low. Maybe try 24. If this number is set too high you may see spiky cpu. The cpu will be managing the queue. linda On 11/11/13 11:37 PM, tclug at freakzilla.com wrote: > Hehe. One external, 8-bay enclosure, using two SATA ports. The ports > go directly to the motherboard - no additional controller. Server > software is Ubuntu 12.10, with ZFS added on from the zfs-native PPA. > > When I was using this as md-software RAID5, I had two disks in each > half of the enclosure. No performance issues. Now this is 8 disks > rather than 4, and raidz2 (so RAID6) rather than RAID5, but still... > hit play on a video and wait 6 seconds for it to start?... that's a > bit... off. No errors except the three checksum errors I've had. > > > On Mon, 11 Nov 2013, Thomas Lunde wrote: > >> >> All of the drives are in a single external enclosure? >> >> How is that enclosure connected to the rest of the PC? USB? (2? 3?) >> eSATA? FireWire? Something else? >> >> If eSATA, then you may be having issues with a port multiplier. >> >> In any case, it's really hard to troubleshoot by guessing. So, if >> you'd like further help to address performance issues, maybe you >> could provide a full hardware and software description of the system. :) >> >> Thomas >> >>> On Nov 11, 2013, at 9:09 PM, tclug at freakzilla.com wrote: >>> >>> No idea what most of what you said is, no (: >>> >>> These are all identical drives, in an external enclosure, so none of >>> it is my own SATA cables. And again, no errors when they were in a >>> software RAID5 (though there were half as many drives) and nothing >>> in the system logs, which is why I am concerned... >>> >>>> On Mon, 11 Nov 2013, Thomas Lunde wrote: >>>> >>>> >>>> Bit flips like this helped me to discover that two of my 10 SATA >>>> cables were marginal. >>>> >>>> Since these are >2T drives, did you do anything with ashift? >>>> Depending on which ZFS implementation you're using, this question >>>> might not make sense? >>>> >>>> An array of drives where some are faking 512 byte sectors and ( >>>> some are really using 512 byte sectors OR some are using 4K sectors >>>> ) can cause abysmal performance. >>>> >>>> Thomas >>>> >>>> _______________________________________________ >>>> TCLUG Mailing List - Minneapolis/St. Paul, Minnesota >>>> tclug-list at mn-linux.org >>>> http://mailman.mn-linux.org/mailman/listinfo/tclug-list >>> _______________________________________________ >>> TCLUG Mailing List - Minneapolis/St. Paul, Minnesota >>> tclug-list at mn-linux.org >>> http://mailman.mn-linux.org/mailman/listinfo/tclug-list >> _______________________________________________ >> TCLUG Mailing List - Minneapolis/St. Paul, Minnesota >> tclug-list at mn-linux.org >> http://mailman.mn-linux.org/mailman/listinfo/tclug-list >> > _______________________________________________ > TCLUG Mailing List - Minneapolis/St. Paul, Minnesota > tclug-list at mn-linux.org > http://mailman.mn-linux.org/mailman/listinfo/tclug-list