On Mon, 20 Sep 2004 21:37, Josh Bonnett <[EMAIL PROTECTED]> wrote: > >Do you have benchmark results to support this assertion? Last time I > > tested the performance of software RAID-1 on Linux I was unable to get > > anywhere near 2x disk speed for writing. > > Not to be a stickler but i hope you mean reading
Correct. > > I did tests by reading two files that were 1G in > >size and the operation took considerably longer than reading a single 1G > > file from a non-RAID system. If RAID-1 was delivering twice the read > > throughput then I should be able to read two 1G files concurrently from a > > RAID-1 in the same time as would be taken to read a single 1G file from a > > single disk. > > also i think that the original poster was assuming that the raid 1 > driver would read in stripes as you would a raid 0, its not required to > implement raid 1 as far as i know, but it might be a nice thing to add > (not sure if it fits nicely in the Linux software raid drivers). You > might have to trade some memory usage and cpu to make sure the blocks > were put back together again(a fifo buffer 2 stripe sizes big would > probably be all it would need) so whether it helps would all come down > to where the bottleneck is. I believe that the Linux software RAID-1 code already spreads the read load between the disks. The problem is that the algorithm chosen is not much good (or at least was not much good last time I tested it). If you had RAID-1 running over flash disks, NVRAM devices or other media that does not have a seek overhead then I expect that read performance would double whenever you have at least two processes reading from the same RAID-1 device. But when seeks matter the algorithm loses. -- http://www.coker.com.au/selinux/ My NSA Security Enhanced Linux packages http://www.coker.com.au/bonnie++/ Bonnie++ hard drive benchmark http://www.coker.com.au/postal/ Postal SMTP/POP benchmark http://www.coker.com.au/~russell/ My home page -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
