I'm checking into this, but I'm strongly suspicious it's a benchmark
anomaly. I'm curious to see what kind of performance you'll get in a
real world application. I rarely pay much attention to benchmarks,
especially when servers are involved, as benchmarks are typically
written for desktops, and if you run them on servers, they don't always
report as they should. In this example, the performance you're getting
is similar (real world) to what I've seen out of 2010 and a number of
Single Channel RAID cards, such as Adaptec 2120 and LSI 320-1 cards.
What I suspect is happening is that since your CPUs aren't really
involved in any work, they've got TONS of resources to devote to your
RAID. As you would suspect, a Dual Xeon system is going to be more
powerful than an $285 RAID card, so when your system isn't busy actually
working, it's got plenty of system resources handy to run your RAID
array VERY fast. However, I'm suspicious that if you were to be running
real world applications, you're probably see much different results.
I've worked on a large number of systems based on the exact same
board/RAID card combination, and a number of other boards that use that
RAID card, and had great luck. I think you're situation is just a
benchmark one, and real world results will be completely different. I'm
doing some further research on this, but those are my preliminary thoughts.
Ed
J. Sokalski wrote:
> The Adaptec 2010S seens to be a bottleneck restricting the speed of my SCSI
> raid arrays?
>
> I am trying to debug and setup an X5DA8 board and have been extremely
> disappointed in the slow SCSI drive speeds when using an Adaptec 2010S in
> the green PCI-X slot. 4 Seagate ST336607LW drives only give 72.2 MB/s
> throughput (across the whole 140GB) in a Raid 0 array. Win2K Server's
> software Raid 0 with the 2010S still in the system gives 104 MB/s over the
> whole 140GB array (a 44% improvement). That indicates there is a SCSI
> bottle neck when a 2010S is in the system. I see the same problem when I
> tried 2010S Raid 5 with the fast Seagates. I observed 54.3MB/s from the
> 2010S and 75.4MB/s with software Raid (a 40% improvement).
>
> Each SCSI 320 drive (according to Seagate's specs.) should provide 78-43
> MB/s throughput from the outside to inside tracks. I expected the outer
> 20GB in a Raid 0 array to show about 300 MS/s. The data comes from WinBench
> 99 v.2's disk inspection tests.
>
> The system board and 2010S have been flashed to the latest versions and
> drivers are the latest versions also.
>
> I don't understand why Supermicro X5DA8 would lose SCSI performance when
> using the recommended Adaptec 2010S raid controller. The performance of the
> hard drives should be better with hardware raid.
>
> A 3Ware 7506-12 controller with 6 IBM 18GXP drives in a Raid 5 array gives
> 192.0 to 152.0 MB/s throughput (across the whole 900GB) so it seems the
> PCI-X slots or motherboard may not be the bottleneck. Also, the Seagate
> Cheatah's specs. are faster than the 180GXP's.
>
> The Supermicro X5DA8 motherboard uses Adaptec's 7902 (Ultra 320) chipset for
> the on board SCSI. It is a dual channel controller. The Adaptec 2010S raid
> controller is plugged into a green PCI-X 64 MHz x 64 bit slot that is
> intended for the 2010S. The PCI-X slots are connected to the high speed bus
> of the motherboard's Intel 7505 chipset.
>
> I tried further tests without the 2010S in the system:
>
> I removed the 2010S controller and ran the 4 Seagate ST336607LW drives as a
> Win2K Server software raid-0 array and observed a huge improvement in
> performance. WinBench 99 v.2 Disk Inspection tests reported 269MB/s on the
> outer track and 161 MB/s on the inner track with 12 steps from high to
> slower performance. With the Adaptec 2010S raid controller the performance
> was a flat line across the disks. The steps are normal when there is no
> bottle neck obstruction their performance.
>
> Has anyone used the 2010S with the Supermicro X5DA8 motherboard and not
> found a big reduction in SCSI performance? Any tip how to make the 2010S
> allow the same throughput as a software raid?
>
> If my 2010S is defective or engineered to be slower than software raid, I
> would like a raid controller that is compatible with the X5DA8's green PCI-X
> slot and is faster than software raid, any suggestions?
>
> Thank you,
>
> Joseph Sokalski
>
><!-- ~MESSAGE_AFTER~ -->
FAQ
Profile
Private Messages
Log in
