Assuming that this problem could be related to a drive firmware bug I have:
1. tuned off NCQ -- or in fact limited the queue depth to 1 2. used truss with the cp command I found this for NCQ: http://blogs.sun.com/erickustarz/entry/ncq_tunable ============================================= NCQ sata_max_queue_depth tunable Previously, i did some analysis on NCQ in OpenSolaris. It turned out that to get good multi-stream sequential read performance, you had to disable NCQ via the tunable 'sata_func_enable'. Disabling NCQ actually does two things: 1) sets the number of concurrent I/Os to 1 and 2) changes what you send down protocol wise. Turns out, the first is all we really need to get good performance for the multi-stream sequential read case, and doing the second actually exposes a bug on certain firmware of certain disks. So i highly recommend the newly added 'sata_max_queue_depth' tunable instead of 'sata_func_enable'. As a reminder, put the following in /etc/system and reboot: set sata:sata_max_queue_depth = 0x1 ============================================= Anyway, after adding the line above into /etc/system, I rebooted and then re-tried the copy with truss: truss cp -r testdir z4 It seems to hang on random files -- so it's not always the same file that it hangs on. On this particular run here are the last few lines of truss output, although they're probably not useful: write(5, " r8A pF2D9B69DCF96ADD2 O".., 8388608) = 8388608 mmap64(0xFE400000, 8388608, PROT_READ, MAP_SHARED|MAP_FIXED, 4, 0x23000000) = 0xFE400000 write(5, "C0A6 IF5 IB9A6 7 0F794 }".., 8388608) = 8388608 mmap64(0xFE400000, 8388608, PROT_READ, MAP_SHARED|MAP_FIXED, 4, 0x23800000) = 0xFE400000 write(5, "8014D819B407 f ?F8F3E5FF".., 8388608) = 8388608 mmap64(0xFE400000, 8388608, PROT_READ, MAP_SHARED|MAP_FIXED, 4, 0x24000000) = 0xFE400000 write(5, " q01DFD0 + % y7FB8 ^ >AD".., 8388608) = 8388608 mmap64(0xFE400000, 8388608, PROT_READ, MAP_SHARED|MAP_FIXED, 4, 0x24800000) = 0xFE400000 Any ideas what I can do next? This message posted from opensolaris.org _______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
