Date: Thu, 09 Nov 2006 13:59:34 +0100 From: Ivan Voras <ivoras@fer.hr> To: Pete French <petefrench@ticketswitch.com> Cc: freebsd-stable@freebsd.org Subject: Re: Dissapointing performance of ciss RAID 0+1 ? Message-ID: <45532636.5000106@fer.hr> In-Reply-To: <E1Gi9OJ-000Atd-1z@dilbert.firstcallgroup.co.uk> References: <E1Gi9OJ-000Atd-1z@dilbert.firstcallgroup.co.uk>
next in thread | previous in thread | raw e-mail | index | archive | help
Pete French wrote: >> It would be interesting for you to track iostat (i.e. run "iostat 1") >> with and without modified vfs.read_max and see if there's a difference. > > On the file: KB/t is about 127.5 with both sizes. Rate is 39 on with > the read_max set to 8, but 115 with read_max set to 64. Ok, this might mean the time has come to increase the default value for vfs.read_max. > On the raw device: KB/t is always 128. rate is 41 with the size set to 8 > but rises to 57 with the size set to 64! How can the vfs parameters affect > access to the raw device ? Don't know. Maybe it's a statistical anomaly (burst)? >> In a similar experiment, you could watch gstat (also before and after) >> and see if it reports the difference. > > On the file: read_max=8 gives 75% busy, 42 meg/sec. 64 gives 99.7% and 120 > On the device: both sizes give the same results - 98% busy, 59 meg/second > > I am not sure this is helping my understanding! :-) That makes two of us :) I think I'll leave this thread to someone with more knowledge of VFS to explain.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?45532636.5000106>