Date: Thu, 29 Jan 2009 23:43:11 -0500 (EST) From: Charles Sprickman <spork@bway.net> To: Richard Tector <richardtector@thekeelecentre.com> Cc: freebsd-stable@freebsd.org Subject: Re: 7.1, mpt and slow writes Message-ID: <alpine.OSX.2.00.0901292253160.37588@toasty.nat.fasttrackmonkey.com> In-Reply-To: <49819672.1090208@thekeelecentre.com> References: <alpine.OSX.2.00.0901290351250.37588@toasty.nat.fasttrackmonkey.com> <49819672.1090208@thekeelecentre.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, 29 Jan 2009, Richard Tector wrote: > Charles Sprickman wrote: >> Hello, >> >> I think this needs a few more eyes: >> >> http://lists.freebsd.org/pipermail/freebsd-scsi/2009-January/003782.html >> >> In short, writes are slow, likely do to the write-cache being enabled on >> the controller. The sysctl used in 6.x to turn the cache off don't seem to >> be in 7.x. > > I have two Dell 860's here also running 7.1-REL. > A simple dd on one shows roughly 60MB/s writes to a mirror of 2 320GB disks > on what I'm assuming is the same LSI controller card (SAS5) > > richard@moses:~# cat /boot/loader.conf > hw.mpt.enable_sata_wc=1 > > Seems to work just fine for me. Well, now I'm puzzled. A "sysctl -a" shows no "hw.mpt." tree at all: # sysctl -a|grep hw.mpt # It's a poor test, but: # dd if=/dev/zero of=/jails/foo count=1M 1048576+0 records in 1048576+0 records out 536870912 bytes transferred in 144.830977 secs (3706879 bytes/sec) # dd if=/jails/foo of=/dev/null 1048576+0 records in 1048576+0 records out 536870912 bytes transferred in 4.136582 secs (129786112 bytes/sec) I added the tunable to loader.conf: $ cat /boot/loader.conf hw.mpt.enable_sata_wc=1 And still nothing in the sysctl output (there was on 6.4 on this same hardware, FWIW): $ sysctl -a |grep hw.mpt $ However, performance did improve: # dd if=/dev/zero of=/jails/foo count=16M 16777216+0 records in 16777216+0 records out 8589934592 bytes transferred in 276.332753 secs (31085474 bytes/sec) Any idea what happened to the sysctl? Is there some other method to verify the loader tunable took (other than testing the throughput)? I assume there's still no management tools for these controllers, right? Thanks, Charles > Regards, > > Richard >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?alpine.OSX.2.00.0901292253160.37588>