Date: Tue, 2 Apr 1996 11:13:54 +0200 (MET DST) From: Luigi Rizzo <luigi@labinfo.iet.unipi.it> To: msmith@atrad.adelaide.edu.au (Michael Smith) Cc: davidg@Root.COM, dutchman@spase.nl, freebsd-hackers@freebsd.org Subject: Re: HDD cpu usage (IDE vs. SCSI). Message-ID: <199604020913.LAA25525@labinfo.iet.unipi.it> In-Reply-To: <199604020133.LAA08679@genesis.atrad.adelaide.edu.au> from "Michael Smith" at Apr 2, 96 11:03:09 am
next in thread | previous in thread | raw e-mail | index | archive | help
DG said: > > > Because a Pentium-90 is much faster than a 486-100 for certain things and > > > most of the %CPU is for total I/O overhead, not just the overhead in the > > > device driver. > > > > Hmmm... if I get it right this means that under certain circumstances > > (1 disk, onboard IDE controller, medium-fast CPU) using SCSI instead > > of IDE gives you only a very little saving (which BTW is what I am > > convinced of, but this has not been the dominating opinion on this > > list). > > No, you're not understanding. For a given CPU, IDE will _always_ use more > CPU time than SCSI. Period. I know I shouldn't have posted this :) I only meant that IF "most of the %CPU is for total I/O overhead, not just the overhead in the device driver" (as DG said) THEN "For a given CPU, IDE will _always_ use (POSSIBLY LITTLE) more CPU time than SCSI. Period." > If you have lots of free CPU, then IDE is fine, but if you feel that your > CPU has better things to do with its time than copy data to and from > your disk, then SCSI is the only solution that makes sense. About this, and the frequent claims that IDE is a no-go for a multiuser system, I would like to remind people a not-too-old message, always from DG, about the 1000 users on ftp.cdrom.com. I noticed that most of user processes were idle, and I thought it was because of lack of disk i/o BW. According to David, the bottleneck was actually in the network bandwidth (mostly on the client side). A single disk doing random seeks can approximately serve some 100 requests/s (parameters derived with bonnie on some of my IDE disks, and I don't think SCSI disks can do much better -- the mechanics is mostly the same, only the controller changes). In a system like ftp.cdrom.com most requests tend to be short -- like 8KB each or so. PIO transfers over a PCI or other built-in IDE controllers will consume 1ms or likely less, and after all you have to compare this with the time necessary to setup your SCSI controller and the remaining (common) FS overhead. Of course, ftp.cdrom.com uses SCSI -- where would you find room for 20+ IDE disks on a system :) Luigi ==================================================================== Luigi Rizzo Dip. di Ingegneria dell'Informazione email: luigi@iet.unipi.it Universita' di Pisa tel: +39-50-568533 via Diotisalvi 2, 56126 PISA (Italy) fax: +39-50-568522 http://www.iet.unipi.it/~luigi/ ====================================================================
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199604020913.LAA25525>