On Tue, 30 Nov 2004, Neil Brown wrote:
On Monday November 29, raid@xxxxxxxx wrote:
now i get 275MB/sec with raid0 (14 disks) 171MB/sec with raid5 (14 disks)
while the old raid5 situation was 257MB/sec with raid5 (13 disks)
I presume you are only looking at 'read' speed here.
yes i'm looking at the read spead
just tested 2.6.10-rc2
142148
this is 50% of the performance i get with the 2.4 kernel (the 171MB/sec i got yesterday with 14 disks was also with xfs) (xfs on 14d r5 gives 171371 read/ext3 on 14 r5 gives 132845 read)
(2.6.10-rc2 13d r5 ext3)
Version 1.03 ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
storage.a2000.nu 2G 24730 99 108265 88 59460 59 25759 91 142148 69 586.8 3
------Sequential Create------ --------Random Create--------
-Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
16 1432 78 +++++ +++ +++++ +++ 1456 77 +++++ +++ 4908 97
storage.a2000.nu,2G,24730,99,108265,88,59460,59,25759,91,142148,69,586.8,3,16,1432,78,+++++,+++,+++++,+++,1456,77,+++++,+++,4908,97
There is a read optimisation that I haven't implemented in 2.6 yet that could account for about a 10% difference, but a 30% difference definitely surprises me.
Normally raid5 will read into the stripe cache, and then copy data out of the stripe cache and into the request buffer for reads. The optimisation is to read directly into the request buffer in situations where no degraded stripes or pending write requests. This optimisation is implemented in 2.4 and gave me about a 10% speedup on sequential reads. Its substantially more complicated in 2.6...
I might do a bit of testing myself, but more results from others would be helpful...
NeilBrown
- 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
