Return to BSD News archive
Path: euryale.cc.adfa.oz.au!newshost.carno.net.au!harbinger.cc.monash.edu.au!msunews!news.mtu.edu!walter.acs.nmu.edu!cloudbreak.rs.itd.umich.edu!newsxfer3.itd.umich.edu!www.nntp.primenet.com!nntp.primenet.com!news.enteract.com!news.he.net!nr1.scn.co.jp!news01.so-net.or.jp!sinfony-news01!sinfony-news02!jun From: jun@goten.sinfony.ad.jp (Junichi Kurokawa) Newsgroups: comp.unix.bsd.freebsd.misc Subject: Re: FreeBSD does interleaved paging : faster? Date: 04 Feb 1997 07:11:05 GMT Organization: Sony Systems Design Corp., Takanawa, Tokyo Lines: 45 Message-ID: <JUN.97Feb4161105@goten.sinfony.ad.jp> References: <JUN.97Feb4100600@goten.sinfony.ad.jp> <32F6BA44.2DA758F0@indiana.edu> NNTP-Posting-Host: goten.gotenyama.sinfony.ad.jp Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII In-reply-to: Lars Hofhansl's message of 04 Feb 1997 13:25:42 +0900 Xref: euryale.cc.adfa.oz.au comp.unix.bsd.freebsd.misc:35008 In article <32F6BA44.2DA758F0@indiana.edu> Lars Hofhansl <lhofhans@indiana.edu> writes: >Junichi Kurokawa wrote: >> >> The subject says it all.... >> >> With dedicated hardwares e.g. Suns, I occasionally hear that interleaved >> swap partitions improve paging performance. >> >> But how about the case with a ubiquitous PC MB plus *signle channel* >> SCSI a la AHA2940, and FreeBSD? Is it faster? >> >> Your comments are welcomed, thanks! >> >> Regards, >> junichi > >It's definately faster if you use seperate _disks_ (assuming all >the disk are about the same speed). Separate across what? Do you mean to divide the swap across the drives, or swap in sd0a and /usr in sd1a or similar as you mention below? >On an older system I had swap and /usr on different disk, which >gave quite a preformance gain. This makes sense. I'll try it. BTW where did you put your /home, a busy partition as well? In the swap drive, the /usr drive or even in a third drive? This is crucial. >Having serveral swappartitions of different disks should give >some performance gain also, since it (in the best case) halfs the >seektime (since the seeks can be done in parallel) as seen by the >operating system (the "page latency" of course is unchanged, and the >throughput may and may not by changed). This was my first intension in the Q. I thank Lars for his comments, and welcome more comments on the subject from others. Regards, junichi