Where did you get 64 and 4? I don't know your fs block size. I assume it is 4k. Your chunk size is 32k. Your stripe size is 32K * (3-1) 1. stride based on chunk size Stride = chunk/block 8 = 32K/4K 2. stride based on stripe size Stride = chunk * (N-1) / block 16 = 32*(3-1) / 4 Please re-do your tests with stride of 8 and 16. Thanks, Guy -----Original Message----- From: John Lange [mailto:john.lange@xxxxxxxxxxxxxx] Sent: Thursday, May 20, 2004 3:59 AM To: Guy Cc: 'LinuxRaid' Subject: RE: Please review: Slackware RAID How-To Hey Guy. Thanks very much for your research. Since we couldn't seem to come up with a definitive answer I decided to resort to some tests. I have a little RAID 5 array comprised of 3 8G SCSI disks in a old Dell PowerEdge (same box mentioned in the How-To). Just for reference here is it's raidtab: raiddev /dev/md0 raid-level 5 nr-raid-disks 3 nr-spare-disks 0 chunk-size 32 persistent-superblock 1 parity-algorithm left-symmetric device /dev/sdb1 raid-disk 0 device /dev/sdc1 raid-disk 1 device /dev/sdd1 raid-disk 2 I used mke2fs to format the drive in 3 different ways with variable stride settings then ran bonnie++ on it. The settings were: stride=64 stride=4 stride=512 The first two are based on the different formulas we discussed. I tried 512 just because I thought I'd do something a bit crazy to see if it made any difference. The results in bonnie++ were almost statistically identical in every case. The only difference in performance was a 20% drop in random seeks when stride=512 but the rest of the performance indicators were almost identical. I repeated the test to make sure it wasn't a blip. So the bottom line is; for me, neither chunk-size OR stride makes any difference to performance. John Lange On Thu, 2004-05-20 at 00:02, Guy wrote: > More info! > > I found this comment in another group. It would indicate "stripe size" > should be used. But it should not make a difference. > > Guy > > ======================================================================= > The stride option places the inode and block bitmaps so that successive > block groups' bitmaps are on a different RAID stripes. I suppose this > might improve disk I/O performance, as the bitmaps are the most heavily > used blocks on the disk. However, the cache should prevent most of the > I/O in the first place... > > -----Original Message----- > From: linux-raid-owner@xxxxxxxxxxxxxxx > [mailto:linux-raid-owner@xxxxxxxxxxxxxxx] On Behalf Of John Lange > Sent: Wednesday, May 19, 2004 11:34 PM > To: office@xxxxxxxxxxxx > Cc: LinuxRaid > Subject: RE: Please review: Slackware RAID How-To > > No, stride is not needed for RAID 1 as stride is about striping and RAID > 1 does not do striping, it does mirroring. > > According to man raidtab, chunk-size "Sets the stripe size to size > kilobytes.". So unless I'm completely off my rocker, chunk-size is also > not needed for RAID 1 as it also only applies to striping. > > Thanks for pointing that out. I have removed it from my HowTo. > > I believe this error is also in the Software RAID HowTo which is where I > copied my examples from. > > Regards, - To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
