Date: Mon, 3 Apr 1995 01:03:08 -0700 (PDT) From: "Rodney W. Grimes" <rgrimes@gndrsh.aac.dev.com> To: taob@gate.sinica.edu.tw (Brian Tao) Cc: freebsd-current@FreeBSD.org Subject: Re: New installation notes Message-ID: <199504030803.BAA03823@gndrsh.aac.dev.com> In-Reply-To: <Pine.BSI.3.91.950403140131.20912k-100000@aries.ibms.sinica.edu.tw> from "Brian Tao" at Apr 3, 95 02:14:36 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> > On Sun, 2 Apr 1995, Rodney W. Grimes wrote: > > > > There is a very big difference between a Quantum Maverick 540 (3600 RPM, > > 14mS, 128k cache) and an Empire 1080 (5400RPM, 9.5mS, 512k cache). The > > two sets of numbers your reported above are very close to the maximum > > the drives can do. > > I only looked up the drive specs this morning, after having > thought Quantum made only 5400 and 7200 drives. That was pretty much > the only explanation I could think of (RPM). Quantum makes 3600, 4500, 5400 and 7200 RPM drives. They have only being making 5400 and 7200 RPM drives for a little over a year (just like every one else). > But I were to do > something like two 'dd if=blah of=/dev/null bs=65536' on files from > each drive simultaneously, I should still be able to hit close to the > maximum throughput, no? Depends on what maximum throughput you are trying to hit. If it is for the SCSI bus/controller a much better test would be to read the same ``disk cache size'' block of data repeatedly from the raw device. repeat 1000 dd if=/dev/rsd0c of=/dev/null bs=size_of_drive_cache count=1 but dd's verbose output slows this way down :-(. I have done dual drive concurrent iozones using 3MB/sec drives and saw a very small difference in the numbers for each drive compared to running them seperately (ie, the bottleneck was really the drives and not the SCSI bus, controller or PCI bus.) This was using a NCR 53C825 based controller running 2 DEC 3053L fast scsi drives (not wide). One needs to be very carefull when doing benchmarks to understand what it is you are really measureing. And to attempt to make the test measure what you really want it to measure. I image a carefully written C program could do the ``repeat 1000 dd'' above and obtain close to SCSI bus bandwidth if that is what you wanted to measure. (I also suspect in this case the real bottleneck would be the disk controller itself). -- Rod Grimes rgrimes@gndrsh.aac.dev.com Accurate Automation Company Custom computers for FreeBSD
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199504030803.BAA03823>