AAC 2200s and firmware 7244

[Andreas Braukmann]

On 05/07/04 21:03:00 -0400 Don Bowman wrote:
> From: Kenneth D. Merry [mailto:ken at kdm.org]

>> On a slightly-related topic, any idea what sort of 
>> performance have you been able to get out of the 2200S?
>> (The number of disks and what kind of
>> array(s) you have would be helpful to know.)

FreeBSD 4.9, Adaptec 2200S with BBU (write back cache)
4 * 36 GByte 10k (Fujitsu) RAID-5

neon# dd if=/dev/zero of=output bs=64k count=200000
200000+0 records in
200000+0 records out
13107200000 bytes transferred in 298.124592 secs (43965511 bytes/sec)

neon# dd of=/dev/null if=output bs=64k count=100000
100000+0 records in
100000+0 records out
6553600000 bytes transferred in 82.796667 secs (79152945 bytes/sec)


>> I'm mainly wondering about sequential performance, since 
>> random performance is harder to gage.

Harder to gage, but most of the time much more interesting;
the performance of massive parallel processes doing sequential
reads or writes are leading to a nearly random access pattern
for the disk system anyways. (see below)

>> e.g., with my Adaptec 5400S/HP NetRAID 4M, I can get about 
>> 92MB sequential read performance with a 6 disk RAID-5, and up
>> to 50MB/sec sequential write performance on the same array.

My HP NetRAID 4M (128 MB Cache, BBU) delivers quite similar
performance. Currently it has only 4 disks and runs under heavy
database load. I can't do sensible tests at the moment.
On single process sequential accesses it delivers similar, but
slightly lower performance than the 2200S.
But: this is on a single tasking sequential access pattern.
The 5400 shines when it comes to heavy duty tasks. In doing
multiple bonnie++ (or postmark) runs in parallel over one or
multiple arrays (or multiple filesystems on the same array)
the 5400 maxes out much later than the 2200S. The 5400 delivers
nearly twice as much i/o-ops per second than the 2200S.


>> With a RAID-1 array composed of the same disks, I can get 
>> 66MB/sec read
>> performance, and up to about 30MB/sec write performance.

The RAID-0 / RAID-1 performance of the higher end RAID
controllers sucks big time. One has to suppose, that they
are heavily optimized to do a good job on RAID-5, leading
to sub-optimal performance for the other RAID levels.
I experimented (5400S) with 6 and 8 disk RAID 5+0 setups;
the performance was worse than a 4 to 6 disk RAID 5 setup.


 
> My experience has been that the ASR (5400s, 2010s, etc)
> outperform the AAC (2200s). 

The 5400S is an aac(4) controller. The 2200S runs circles
around the lower end asr(4) controllers. I don't have any
experience with the higher end asr(4) controllers (Adaptec
3200 or 3400), though.

For sequential access patterns the 3Ware Controllers are
running circles around the scsi crowd.

This is an 3Ware 7506 writing to a 4 disk (Maxtor 250 GB,
7200 UPM) RAID-5 volume:

dagobert# dd if=/dev/zero of=output bs=64k count=200000
200000+0 records in
200000+0 records out
13107200000 bytes transferred in 264.988240 secs (49463327 bytes/sec)

And now it's reading:
dagobert# dd if=output of=/dev/null bs=64k count=100000
100000+0 records in
100000+0 records out
6553600000 bytes transferred in 58.935684 secs (111199185 bytes/sec)


I'm eager to test the Escalade 9000 series with cache, BBU
and a few WD Raptor drives.


-Andreas