Date: Fri, 22 Nov 1996 18:29:54 +1100 (EST) From: michael butler <imb@scgt.oz.au> To: freebsd@dyslexic.phoenix.net (FreeBSD Acct) Cc: hackers@freebsd.org Subject: Re: Disk Striping Message-ID: <199611220729.SAA11415@asstdc.scgt.oz.au> In-Reply-To: <Pine.BSF.3.91.961121233424.5121C-100000@dyslexic.phoenix.net> from FreeBSD Acct at "Nov 21, 96 11:37:55 pm"
next in thread | previous in thread | raw e-mail | index | archive | help
> Our news servers here, are P166's with 128Mb each, with two Adaptec3940 > PCI cards. > There are four 4Gig drives, each has its own separage SCSI controller as > the 3940 has two controllers per card. > The drives are identical, and in a CCD array. [ .. ] > Compared to how we had the news farm before..the CCD is many many times > faster and more efficient. With no readers (and therefore no need for .overview files all over the place to mess things up), this combo will *scream* along quite happily at better than 10 articles a second iff the accesses for history and articles are on separate CCD arrays on separate channels. Short of getting "async" mode to work properly, there's not much better. The general principle is .. Seeks cost a lot, which can be relieved by configuring drives (amalgamated or not) to operate on only one part of the data set (history or articles but not both). Rotational latency costs less and can be divided by distributing across N drives in a CCD array, michael
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199611220729.SAA11415>