Date: Fri, 12 Jan 2007 02:30:04 -0700 From: Scott Long <scottl@samsco.org> To: Ahnjoan Amous <ahnjoan@gmail.com> Cc: freebsd-fs@freebsd.org Subject: Re: aac scsi raid driver performance Message-ID: <45A7551C.5030006@samsco.org> In-Reply-To: <5e575c8a0701111307s5a839b82ra4ba5c45d554d3b9@mail.gmail.com> References: <5e575c8a0701111155l859d0ecif617dbda43cef842@mail.gmail.com> <45A6A138.8090208@samsco.org> <5e575c8a0701111307s5a839b82ra4ba5c45d554d3b9@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Turn off read caching Ahnjoan Amous wrote: > I reset the controller to the defaults for everything before I did the > installation. > > Each of the containers were created with the following options. > Container Type - RAID 0 > Container Label - data03 > Container Size - 279.396 GB > Chunk Size - 64KB > Read Caching (Yes/No) - Y > Write Caching : Enable when protected > Create RAID 5 via - N/A > > Then the controller itself has a single cache option > Drives Write Cache - Disabled > > Thanks > Ahnjoan > > On 1/11/07, Scott Long <scottl@samsco.org> wrote: >> Did you enable read caching for the arrays? >> >> Scott >> >> >> Ahnjoan Amous wrote: >> > I'm trying to find possible explanations for slow concurrent writes >> > through the >> > aac driver. This machine runs under 1% load and has less than 4 >> > transfers per >> > second to the drives in question when not being used for testing. >> > >> > When I attempt sequential "dd"s as follow, the results are better then >> > 70MB/sec. >> > dd if=/dev/zero of=/data02/helloworld bs=1m count=1000 >> > 1048576000 bytes transferred in 13.886718 secs (75509274 bytes/sec) >> > dd if=/dev/zero of=/data03/helloworld bs=1m count=1000 >> > 1048576000 bytes transferred in 14.011323 secs (74837758 bytes/sec) >> > >> > When I attempt concurrent "dd"s as follow, with a 1 second sleep >> interval >> > between starts, the results are better than 40MB/sec >> > dd if=/dev/zero of=/data02/helloworld bs=1m count=1000 & >> > sleep 1 >> > dd if=/dev/zero of=/data03/helloworld bs=1m count=1000 & >> > 1048576000 bytes transferred in 25.269555 secs (41495626 bytes/sec) >> > 1048576000 bytes transferred in 24.935765 secs (42051086 bytes/sec) >> > >> > When I attempt concurrent "dd"s as follow, the results are little >> better >> > than >> > 20MB/sec >> > dd if=/dev/zero of=/data02/helloworld bs=1m count=1000 & >> > dd if=/dev/zero of=/data03/helloworld bs=1m count=1000 & >> > 1048576000 bytes transferred in 44.963408 secs (23320652 bytes/sec) >> > 1048576000 bytes transferred in 45.010065 secs (23296478 bytes/sec) >> > >> > I can't account for what causes the huge difference however the >> results are >> > reproducible. I've run the tests dozens and dozens of times now, first >> > blaming >> > the em driver for my slow ggatec/ggated results, then GEOM for my slow >> > local >> > mirroring after eliminating the network, and finally blaming the aac >> driver >> > after removing GEOM from the equation. If anyone has ideas on what >> I might >> > look at or change or test I would love to hear. >> > >> > >> > ****** Misc. Information ****** >> > root:somehost:~ > df -k >> > Filesystem 1K-blocks Used Avail Capacity Mounted on >> > /dev/aacd2s1e 2026030 1024532 839416 55% /data02 >> > /dev/aacd3s1e 2026030 1024532 839416 55% /data03 >> > Hardware - >> > aac - Dell PERC3/Di U160 >> > aacd2 - hardware RAID 0, w/1 U320 300G drive >> > aacd3 - hardware RAID 0, w/1 U320 300G drive >> > _______________________________________________ >> > freebsd-fs@freebsd.org mailing list >> > http://lists.freebsd.org/mailman/listinfo/freebsd-fs >> > To unsubscribe, send any mail to "freebsd-fs-unsubscribe@freebsd.org" >> >>
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?45A7551C.5030006>