Date: Tue, 26 Oct 1999 17:09:34 +0100 From: Geoff Buckingham <geoffb@chuggalug.clues.com> To: Greg Lehey <grog@lemis.com> Cc: Mike Tancsa <mike@sentex.net>, freebsd-current@FreeBSD.ORG Subject: Re: AMRD (MegaRAID) BIOS rev another other questions Message-ID: <19991026170934.B95560@chuggalug.clues.com> In-Reply-To: <19991026110615.51265@mojave.worldwide.lemis.com>; from Greg Lehey on Tue, Oct 26, 1999 at 11:06:15AM -0400 References: <3.0.5.32.19991020143339.016e0210@staff.sentex.ca> <19991022140949.A82396@chuggalug.clues.com> <19991024095811.61238@mojave.worldwide.lemis.com> <19991026112647.B95499@chuggalug.clues.com> <19991026110615.51265@mojave.worldwide.lemis.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Oct 26, 1999 at 11:06:15AM -0400, Greg Lehey wrote: > On Tuesday, 26 October 1999 at 11:26:47 +0100, Geoff Buckingham wrote: > > On Sun, Oct 24, 1999 at 09:58:11AM -0600, Greg Lehey wrote: > >> > >> Indeed. It's quite easy to put all cylinder groups on a single > >> spindle; I've seen reports of up to 80% degradation under these > >> circumstances. > >> > >>> Where sequential read/write performance is not critical you can stripe at > >>> cluster size to avoid this. Other wise using an odd number of spindles for > >>> a stripe and an even number for a RAID3 or RAID5 or stripeing at an interval > >>> which is not a power of two should work (12,24,48,76 etc) > >> > >> The best I've heard of is 768 kB - 1 sector. This works on Vinum, but > >> it seems that most RAID controllers use a somewhat simplistic striping > >> algorithm. You might like to try 31 kB or such. This won't make any > >> difference with rawio, though. > > > > I would agree from experiance that 768 KB seems a good selection for vinum > > and probably ccd too > > There's a big difference between 768 kB and 767½ kB. That's the point > I was trying to make. With 768 kB, you risk having all your > superblocks on one drive. After having first noticed the problem I initially striped at cylinder group size to avoid it (usually 32MB) and would generally recomend this for entirly random enviroments, even though this wastes some space (where 0<= some < 32MB) per disk and reduces sequential throughput. THis also retains any benefit that the original UFS cylinder group concept was intended to bring. I have only dome a little testing with 768KB stripes but what I did seemed free of the problem. I was looking for a stripe size that wasn't going to cleanly divide into the cylinder group while still being divisable by block size (this feels like the correct thing to do, i have nothing to back up this aversion to spliiting blocks over two disks). That said 760 or 776 may better suit as they would put the begining of fewer cylinder groups on the boundries betweeen stripes. However 768 should work so long as you dont use newfs to intentionally set your cylinder group to be a multiple of 768 kB. newfs could presumably be used to work round potential problems with hardware RAIDS that only offer 16k 32k etc however I am a little nervous about deviating from newfs defaults (as 99.9% of people don't and I don't enjoy being in a minority) Is there a File Systems list we should be having this conversation on? -- GeoffB > > Greg > -- > Finger grog@lemis.com for PGP public key > See complete headers for address and phone numbers To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19991026170934.B95560>