<html><head><meta http-equiv="Content-Type" content="text/html charset=utf-8"></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space;" class="">Chip:<div class="">I tried running fio on multiple folders and it’s a little better<div class=""><br class=""></div><div class="">When I run 7x fio (iodepth=4 threads=4), I get 28k IOPS on average</div><div class="">When I run 7x fio (iodepth=4 threads=16), I get 35k IOPS on average. iostat shows that there is transfer rate of 140-220MB/s with average request size of 35kB</div><div class="">When I run 7x fio (iodepth=1 threads=1), I get 24k IOPS on average</div><div class=""><br class=""></div><div class="">There are still at least 10k IOPS left to use I guess:)</div><div class=""><br class=""></div><div class="">Bob:</div><div class="">Yes, my ZFS is ashift=12, since all my drives report 4k blocks (is that what you ment?). The pool is completly empty, so there is enough place for writes, so write speed should not be limited because of COW. Looking at iostat, there are no reads on the drives at all.</div><div class="">I’m not sure where fio gets its data, probably from /dev/zero or somewhere?</div><div class=""><br class=""></div><div class="">I will try sync engine instead of solarisaio to see if there is any difference.</div><div class=""><br class=""></div><div class="">I don’t have compression enabled, since I want to test raw performance. I also disabled ARC (primarycache=metadata), just so my read tests are also as real as possible (so I don’t need to run tests with 1TB test file).</div><div class=""><br class=""></div><div class=""><blockquote type="cite" class="">Try umounting and re-mounting your zfs filesystem (or 'zfs destroy' followed by 'zfs create') to see how performance differs on a freshly mounted filesystem. The zfs ARC caching will be purged when the filesystem is unmounted.</blockquote></div><div class=""><br class=""></div><div class="">If I understand you correctly, you are saying I should destroy my folders, set recordsize=4k on my pool and then create zfs folders?</div><div class=""><br class=""></div><div class="">thanks, Matej</div><div class=""><br class=""></div><div class=""><br class=""><div><blockquote type="cite" class=""><div class="">On 22 Oct 2015, at 21:47, Schweiss, Chip <<a href="mailto:chip@innovates.com" class="">chip@innovates.com</a>> wrote:</div><br class="Apple-interchange-newline"><div class=""><div dir="ltr" class=""><div class="">The ZIL on log devices suffer a bit from not filling queues well. In order to get the queues to fill more, try running your test to several zfs folders on the pool simultaneously and measure your total I/O. <br class=""><br class="">As I understand it, ff you're writing to only one zfs folder, your queue depth will stay at 1 on the log device and you be come latency bound.<br class=""><br class=""></div>-Chip<br class=""></div><div class="gmail_extra"><br class=""><div class="gmail_quote">On Thu, Oct 22, 2015 at 2:02 PM, Matej Zerovnik <span dir="ltr" class=""><<a href="mailto:matej@zunaj.si" target="_blank" class="">matej@zunaj.si</a>></span> wrote:<br class=""><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Hello,<br class=""> <br class=""> I'm building a new system and I'm having a bit of a performance problem. Well, its either that or I'm not getting the whole ZIL idea:)<br class=""> <br class=""> My system is following:<br class=""> - IBM xServer 3550 M4 server (dual CPU with 160GB memory)<br class=""> - LSI 9207 HBA (P19 firmware)<br class=""> - Supermicro JBOD with SAS expander<br class=""> - 4TB SAS3 drives<br class=""> - ZeusRAM for ZIL<br class=""> - LTS Omnios (all patches applied)<br class=""> <br class=""> If I benchmark ZeusRAM on its own with random 4k sync writes, I can get 48k IOPS out of it, no problem there.<br class=""> <br class=""> If I create a new raidz2 pool with 10 hard drives, mirrored ZeusRAMs for ZIL and set sync=always, I can only squeeze 14k IOPS out of the system.<br class=""> Is that normal or should I be getting 48k IOPS on the 2nd pool as well, since this is the performance ZeusRAM can deliver?<br class=""> <br class=""> I'm testing with fio:<br class=""> fio --filename=/pool0/test01 --size=5g --rw=randwrite --refill_buffers --norandommap --randrepeat=0 --ioengine=solarisaio --bs=4k --iodepth=16 --numjobs=16 --runtime=60 --group_reporting --name=4ktest<br class=""> <br class=""> thanks, Matej<br class="">_______________________________________________<br class=""> OmniOS-discuss mailing list<br class=""> <a href="mailto:OmniOS-discuss@lists.omniti.com" class="">OmniOS-discuss@lists.omniti.com</a><br class=""> <a href="http://lists.omniti.com/mailman/listinfo/omnios-discuss" rel="noreferrer" target="_blank" class="">http://lists.omniti.com/mailman/listinfo/omnios-discuss</a><br class=""> <br class=""></blockquote></div><br class=""></div> </div></blockquote></div><br class=""></div></div></body></html>