Adaptec 5805 SAS slower than expected

[hhansard]

Experienced IT professional, new to passmark and these forums

W7x64 latest patches, AVI disabled, etc

Adaptec 5805Z PCIx8 SAS array 4x450x15K 3G/sec drives configured RAID 0

GigaByte X58-extreme mb

Array should be in the 600 Mb/sec range and 500 Mb/sec sustained, but “feels” much slower than that.

Got passmark to verify my gut. P7 gives numbers in the 190 – 200 read/write range.

OK, I have numbers consistent with my feeling, but can’t for the life of me figure-out what an otherwise stable high-performance array should run like a dog.

Array diags show AOK, write-back is enabled, no problems what so ever.

Should be noted: used to have a FC controller in the same slot some time ago, and got the 650Mb/sec sustained that I wanted. So I know it can be done.

Any ideas?

Thanks in advance, H

[David (PassMark)]

You should check you have the latest device drivers for the RAID controller, defag the disks. If you wanted to take them out of the RAID set you could also look at the SMART values to see if there is a fault developing.

You might also want to try using the advanced disk tests in PerformanceTest to try a few different test scenarios.

We have also seen reports that disabling command queuing in Windows can help with RAID controllers.
Device manager - SCSI and RAID controllers and unchecked the command queuing box for all drives. (might need a restart after this, not sure)

Also I know it isn't what you wanted to hear, but after past grief and poor manufacturer support with hardware RAID controllers I switched to using software RAID.

[hhansard]

Thanks for the reply.

Generally, you can’t defrag through an array unless the software is aware of the topology implementation. At present, I know of no PC-based systems that have this capability. Some SAN and channelized I/O systems can do this, but they are expensive.

Latest board firmware.

SMART is AOK.

Seagate tools report AOK.

I am doing uncompressed 2K video work. There is no way a software raid could keep up. Average video stream is 100GB.
I'll look at the advanced P7 tools.
I’ll look at W7 settings. I’ll also boot PE to see what the P7 boot tools tell me. I have 3 arrays on this box and P7 reports reasonable results for the other controllers, just this array looks funky.

Thanks, H

[David (PassMark)]

There is no way a software raid could keep up.

You might be suprised.

Benchmarks here,
http://www.servethehome.com/windows-dynamic-disk-versus-ich10r-raid-0-intel-x25-m-g2-80gb/
"Windows software RAID does look very strong in comparison to the ICH10R RAID...."

http://www.overclockers.com.au/article.php?id=179581&P=3
"We resorted therefore to Dynamic Disk with Windows XP. This allowed us to combine the 2 Raptors connected to the ICH5-R and the other 2 Raptors connected to the SiL3122A to a striped (RAID 0) array comprising 4 disks on 4 separate channels. Did it work? Oh yes! It did work very well....."

[David (PassMark)]

Generally, you can’t defrag through an array

I am not so sure about this. The fragmentation problem must still occur with stripped disks, and regardless of if the data is spread across multiple physical drives there must be a benefit in getting the file's sectors into sequential order.

All the research I have done would indicate the defagmenting RAID makes sense. e.g.
http://www.diskeeper.com/defrag/file-fragmentation-sans-nas-raid.aspx

[hhansard]

I am. Been using Diskeeper since the program over 10 years. It absolutely does not see below the abstraction layer. Any performance would be a matter of luck only… “[FONT='Arial','sans-serif']more likely, depending on the RAID, to fill the entire 64K chunk…”[/FONT]. I used to speak to Diskeeper tec support years ago on these specific issues. Just give it a try… Actual MB/sec on file-level read/write access will not change appreciably on a decent controller with a good-sized cache. On your typical $100 controller, who knows? Most raid manufacturers specifically recommend against defrag software unless written against their topology implementations.

[David (PassMark)]

Most raid manufacturers specifically recommend against defrag software

Adaptec seem to say it is OK to defrag in their support pages.

[rtitan]

My 5805 with (6) WD2502ABYS SATA II drives in RAID 0...

Disk Mark = 4697.4
Sequential Read = 571.4
Sequential Write = 603.0
Random Seek + RW = 124.4

NCQ is enabled and actually I have background consistency check enabled over 30 days (has a small effect on perfomance).