On Tue, 2012-07-10 at 17:37 +0200, freaky wrote: > On 10.07.2012 08:00, Christoph Hellwig wrote: > > On Mon, Jul 09, 2012 at 11:46:08AM -0700, Andy Grover wrote: > >> It looks like it only wants to allow the use of TYPE_DISK block > >> devices > >> for iblock. Is there another way from userspace to know if a block > >> device is a disk vs. tape or optical? > > > > iblock should not care - it just submits bios. And pscsci already > > does > > it's own internal checks. > > Whilst on the subject userspace, I have exported it with fileio because > I need buffered mode. I know it's a risk with data consistency, but I > need the additional performance (15MB/s writes (~30MB/s via iblock btw - > but that has no cache options whatsoever) and nearly 100MB/s with > buffered (until the buffers are full - MD RAID-6 is performing horribly > (for writes, reads are fine), local as well btw dd'ing zero's with 1MB > block size does somewhere between 20-30MB/s). 8 disk RAID-6 probably has > too much parity overhead. Converting it to RAID-5 but it's gonna take 8 > days or so. Btw, it sounds like your MD raid456 setup is issuing non stripe aligned I/O, which ends up killing any type of reasonable large block I/O performance with software raid pretty quickly. I'd recommend checking your /sys/block/$MD_DEV/queue/max_sectors_kb, and make sure it's aligned properly on stripe boundry in order to do full stripe reads + writes. Also you should try to get a performance baseline with local MD block device performance vs /loopback local SCSI LUN vs. /iscsi remote export, which should help you isolate down the possible bottleneck(s). -- To unsubscribe from this list: send the line "unsubscribe target-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
