Date: Wed, 28 Oct 1998 01:21:42 -0600 From: Dan Nelson <dnelson@emsphone.com> To: Joe Greco <jgreco@solaria.sol.net>, david@sparks.net Cc: hackers@FreeBSD.ORG, mlnn4@oaks.com.au Subject: Re: Multi-terabyte disk farm Message-ID: <19981028012142.A18128@emsphone.com> In-Reply-To: <199810280500.XAA10438@aurora.sol.net>; from "Joe Greco" on Tue Oct 27 23:00:28 GMT 1998 References: <Pine.BSI.3.96.981027231302.23783B-100000@sparks.net> <199810280500.XAA10438@aurora.sol.net>
next in thread | previous in thread | raw e-mail | index | archive | help
In the last episode (Oct 27), Joe Greco said: > Writing to a RAID involves a performance penalty where all drives > have to participate in the write process. Involving only five drives > instead of eight implies that the other three can be doing something > else. That doesn't make much sense. Writing to a RAID should only involve two drives; the data drive and the drive the parity is on. You read the original parity, read the original data, XOR it out of the parity area, XOR the new data back in, and write the updated data and parity (so 2r + 2w for every write the OS makes). Writes larger than the stripe size should access the multiple data+parity drives all in parallel. You pay a constant penalty for writes when going to RAID5, but the number of drives involved in the array doesn't affect it. You only have to read every drive when one fails and you have to recalculate a drive's contents from the parity data. We have four 27+1 x 9GB external RAID5 boxes that take over 4 hours to recalc when a drive fails. -Dan Nelson dnelson@emsphone.com To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19981028012142.A18128>