Re: Perc 5/i SAS

[Havard Eidnes <he%uninett.no@localhost>]

>> dd if=/dev/zero of=bigfile bs=1m count=1000
>> [...]
>> 1048576000 bytes transferred in 11.907 secs (88063828 bytes/sec)
>> [...]
>> mfi0 at pci1 dev 0 function 0: SAS 9260-8i
>> mfi0: interrupting at ioapic0 pin 18
>> mfi0: logical drives 3, version 12.12.0-0111, 512MB RAM
>
> Compared to the adapter RAM, this is not a large write. As long as
> it fits into the cache (provided that the battery is OK), I'm
> getting with an identical adapter, and also 6.0_BETA2:
> 
> # dd if=/dev/zero of=/mnt/bigfile bs=1m count=300
> [...]
> 314572800 bytes transferred in 0.604 secs (520815894 bytes/sec)

Hm, I'm envious, I only get 128MB/s in an identical test.

Anyway, I re-did the test with mor data, and essentially got the
same result:

20971520000 bytes transferred in 240.888 secs (87059214 bytes/sec)

This is with two 1T SATA2 disks in a RAID1 configuration.

Now, I have a couple other logical units in this box, among them
a pair of SATA2 SSDs, also in a RAID1 configuration, but the
performance is most disappointing:

10485760000 bytes transferred in 128.566 secs (81559354 bytes/sec)

I.e. *slower* than the rotating 1TB disks -- it had a sprint at
the very start where it peaked at around 192MB/s according to
"systat vm", but it quickly settled at around 80MB/s.

> The SAS performance is really disappointing. Perhaps OpenBSD
> known which knob to turn to get more out of it. Raw handshake
> and DMA performance don't seem to be the limitation.

I think there must be more than just SAS vs. SATA which is the
problem here.  I think there must be something fishy with the
driver (say I without having taken a closer look at it).

A third logical unit in the box has been mistakenly configured as
a raid6 (it should have been raid10, something I'll be fixing),
and there I get

1048576000 bytes transferred in 32.567 secs (32197500 bytes/sec)

Not much to cheer about that either.

Regards,

- Håvard