Poor write performance with LSI 320-2 on 6.1-STABLE

[Dan Nelson]

In the last episode (Sep 29), Albert Chin said:
> On Thu, Sep 28, 2006 at 05:15:05PM -0500, Albert Chin wrote:
> > I have an Intel S875PWP1 motherboard with a Pentium4 CPU at 2.40GHz. PCI
> > bus is 33Mhz, 32-bit. I recently purchased an LSI 320-2/128MB on eBay
> > (though the card really looks like a PERC4/DS) and just ran some
> > bonnie++ tests on a RAID 1 array between two U320 drives for the first
> > channel and on a RAID 0 array between one U320 drive for the second
> > channel. The 320-2 has the latest LSI firmware, 1L47.
> 
> I reran some of the tests with the same 320-2 but on an Intel
> SE7520BD2 with 32-bit and 64-bit (100Mhz) slots:
>   #1. RAID 1, two U320 drives, channel 1, 32-bit, 33Mhz slot
> Version 1.93c       ------Sequential Output------ --Sequential Input- --Random-
> Concurrency   1     -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
> maetel.il.thew 300M   186  99 16707   5 16063   6   654  99 537320  93  4129  50
> Latency             45215us     199ms   89764us   34740us    1215us    1808ms
> Version 1.93c       ------Sequential Create------ --------Random Create--------
> maetel.il.thewritte -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
>               files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
>                  16  7441  23 +++++ +++ +++++ +++  5799  18 +++++ +++ +++++ +++
> Latency               479ms     122us    2508us     606ms   13549us     101us
> 
>   #2. RAID 1, two U320 drives, channel 1, 64-bit, 100Mhz slot
> Version 1.93c       ------Sequential Output------ --Sequential Input- --Random-
> Concurrency   1     -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
> maetel.il.thew 300M   186  99 18006   6 15964   5   634  99 571275  99  4450  57
> Latency             44992us     139ms     130ms   35143us    1238us     120ms
> Version 1.93c       ------Sequential Create------ --------Random Create--------
> maetel.il.thewritte -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
>               files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
>                  16  7581  24 +++++ +++ +++++ +++  5750  18 +++++ +++ +++++ +++
> Latency               511ms     255us    2615us     622ms   12691us      53us
> 
> Odd that I don't get x2 the performance when the bus bandwidth doubles
> in speed.

Not really odd, since you're nowhere near even the 32-bit bus's max.
(32bit * 33Mhz) / 8 bits = 132 MB/sec, and in write-through mode you're
spending most of your time witing for the disks to sync.  With a larger
filesize you might see a difference in the sequential input test;
judging by your insane sequential read and random seek values, your
300M test file looks like it's completely cached in RAM.  A size 2x
your RAM capacity is recommended.
 
-- 
	Dan Nelson
	dnelson at allantgroup.com