SATA Raid Works!

[eduncan911]

...just on-board RAID.

(copy from another post, but thought it needed it's own thread started here for people to reply):

Just ran HDTach. It's better then normal Raptor drives, but I'm a bit disappointed. Wanted higher rates since I used other programs on my Ultra160s in the past and gotten higher numbers. I went to SATA knowing I'd get a decrease in performance.

HDTach 3.0.1.0
Volume: Promise 2+0 Stripe/RAID0 1.10
Random access: 7.8ms
CPU Utilization: 16%
Average read: 92.1MB/s
Burst Rate: 129.2MB/s


There's proof that SATA RAID 0 works I guess. Then again, I am not using on-board raid:

- Promise SX4 S150 SATA Raid PCI Controller card - 4 port
- Samsung 256MB PC166 ECC memory stick (for write cache on the SATA card!)
- 2x36GB Raptor drives in RAID 0
- 2x200GB Maxtor 7.2k 8MB SATA drives in RAID 1
- System is overclocked currently with 230mhz bus (running the Northbridge 15% over normal, no PCI-lock)

I guess I have a decent system. Makes me wonder what I'd hit with 2x74GBs Raptors... Or better yet, 4x74GBs in RAID 0!

 

[eduncan911]

I went to the expense of using a seperate RAID card, even though my motherboard has two (yes, count them TWO) seperate SATA RAID controllers on-board.

Once is by Intel, and the other by SI (Silicon Image). If I recall, the Intel sata controller is bios/software raid (requires XP drivers, obviously).

But the Silicon Image controller is "hardware". Enabling that, I am able to configure arrays on bootup before POST. I wish I knew that before I went through the expense of the add-on PCI hardware card. Given write access is much faster with the cached add-on card, $130 for the add-on card could have been saved for the lack of cached write speeds.

The other reason I did the add-on card was: upgradability. The Promise card allows me to take the RAID0 drives and move them to another Promise card, or system all togather, without having to rebuild the array! And I change motherboards ever so often.

So I was afraid of having a RAID array (especially the highly organized, a decade worth of work and storage, on the two 200GB drives I have now). I wanted assurance I could move those drives without loosing data. I know, overly cautions. But, I consider the $130 insurance tax to make sure I could do this, since I couldn't get a straight answer from Asus or Abit on moving drives from motherboards to motherboards (system to system), keeping data and arrays intact. The extra performance of the write-back cache is just a bonus.

Oh well, I'm happy (for now).

 

[eduncan911]

Ok. Changing to the "Quick Test" feature, burst rate was the same. But average read jumped to 109.8MB/s.

Someone jsut reported that he hit 226MB/s with his 73GB Raptors... If that's the case, I got some serious tweaking to do... I should be a bit closer thent hat.

I've noticed most people around 2 to 4% CPU utilization. I'm at 16%. Guess I need to fix that...

 

[obiwan]

Not sure if it makes any dif but here's mine.
2 74gb rappys in raid0
Asus P5GDC-v
Intel onboard 82801FR

Burst Speed= 226.6 MB/s
Random Access= 8.0 ms
CPU utilization= 4%
Avg Read= 119.8 MB/s

 

[eduncan911]

Not sure if it makes any dif but here's mine.
2 74gb rappys in raid0
Asus P5GDC-v
Intel onboard 82801FR

Burst Speed= 226.6 MB/s
Random Access= 8.0 ms
CPU utilization= 4%
Avg Read= 119.8 MB/s

Yep, I got some tweaking to do...

Awesome man!

 

[Zac42]

Isn't the PCI bus limited to 133 MB/sec? I dont see how you could any higher burst rate period with a PCI controller card... You would have to have a PCI-E or PCI-X card for more bandwidth. You could get a higher avg read, I guess, but that is still close to completely saturating your PCI bus.

 

[eduncan911]

How about over 1000MB/s burst?

Found this nifty article... Appearently Asus and Abit isn't optimized.

http://members.shaw.ca/xtremecomputing/Review01.htm

I'll believe that, since I've seen > 500MB/s substained (avg read these days) on older SCSI arrays.

 

[Sonny]

You have a 32/64 PCI 2.2 card on the PCI Express P5AD2?

 

[eduncan911]

You have a 32/64 PCI 2.2 card on the PCI Express P5AD2?

Yes, I have the Promise SX4 32/64 PCI 2.2 card. But it's not in a PCI Express slot. It's in a normal 32bit PCI slot.

I'd like to get a good SATA Raid controller card, with cache, for PCI Express 1x. But the ones I found hadn't made it to the US yet. That, and they are like $400. PCI Express seems to be a premium thing right now. I'll wait a little bit before switching to true PCI Express.

 

[Xaotic]

How about over 1000MB/s burst?

Found this nifty article... Appearently Asus and Abit isn't optimized.

http://members.shaw.ca/xtremecomputing/Review01.htm

I'll believe that, since I've seen > 500MB/s substained (avg read these days) on older SCSI arrays.

Even if both drives were being tested only for bursting from the cache, the SATA interface would be limited to less than 150MB/s per channel. This excludes inherent PCI bus bandwidth consideration. So for a buffer to controller transfer, the theoretical maximum is less than 300MB/s. I'm not sure how he was getting his data, but there are some logical gaps in interpreting the results.

 

[eduncan911]

Even if both drives were being tested only for bursting from the cache, the SATA interface would be limited to less than 150MB/s per channel. This excludes inherent PCI bus bandwidth consideration. So for a buffer to controller transfer, the theoretical maximum is less than 300MB/s. I'm not sure how he was getting his data, but there are some logical gaps in interpreting the results.

I'll agree with that. But he's not the only one (look at the submissions lately to HDTach!). Dang... And most of them are using the onboard Intel RAID! Greatly exceeding that 150 limit.

 

[Sonny]

Yes, I have the Promise SX4 32/64 PCI 2.2 card. But it's not in a PCI Express slot. It's in a normal 32bit PCI slot.

So your willing to believe that even with a maximum limitation of 133MB/s on the PCI slot, not counting the 150MB/s X 2 because of RAID0 on the card, you can somehow surpass this dual 74G Raptors. The tests results linked to are not what you would call realistic much less reliable


I don't think Xaotic was saying a 150MB/s total limit but per channel which would make a combined 300MB/s when some people test with Native SATA RAID Controllers that usually come in pairs. The current submission of "sky high" test figures could also mean a problem with the software that everyone is using especially when hardware itself is not capable of it.

 

[Xaotic]

Both HDTach and Sandra have well known issues with RAID systems. Sandra is largely inaccurate for even single drive testing.

On your system, you would get better performance by using the ICH6R connections to isolate the RAID from the PCI bandwidth issue. I'd doubt that any tweaks would significantly improve your performance.

 

[eduncan911]

Good tips about the software.

Yeah, I doubted the tests. What I'd like to see though is HDTach measure my write speeds. Since the cached memory is for write-behind only.

 

[eduncan911]

So your willing to believe that even with a maximum limitation of 133MB/s on the PCI slot, not counting the 150MB/s X 2 because of RAID0 on the card, you can somehow surpass this dual 74G Raptors. The tests results linked to are not what you would call realistic much less reliable

Never said I believed it. Just others seem to be recording these speeds.

I don't think Xaotic was saying a 150MB/s total limit but per channel which would make a combined 300MB/s when some people test with Native SATA RAID Controllers that usually come in pairs. The current submission of "sky high" test figures could also mean a problem with the software that everyone is using especially when hardware itself is not capable of it.

Yeah, I wanted PCI64 slots. But I gave them up in my dual Xeon system when I went to this LGA775. So I lost the bandwidth.

 

[Kamakzie]

Yeah Sisoft sucks, it told me my single Raptor was 12ms and 54mb/s!