Quad Raptor RAID0 not up-to-snuff Options

[wiinter]

Forum-goers – thanks for taking a look at this.

I’m running a four-drive RAID0 setup, all WD Raptor 36gb drives, and it is not performing any better than the benchmarks found online, or my other Raid0 (two-drive) on the same system.

System specs:
• Asus P5N32-E SLI, with Intel E6600, 2gb Corsair XMS2
• RocketRaid 2310 (PCI-e)
• 4x WD Raptor 36gb SATA (two WD360GD, and two WD360ADFD)
• 2x Maxtor 300gb SATA
• 500W Antec Phantom PSU

The four Raptors are running off the RocketRaid. The two Maxtors are running off the Asus SATA controller.

After I had the rig all set up, I fully expected to see some average speeds of 200+ mb/s as I’ve seen all over the web on quad RAID0. I especially expected to see it out perform the two Maxtor drives. But as it stands, here are the HD Tach results:

nVidia controller setup (2x Maxtor RAID0)..
http://www.wiinter.com/pics/HDTACH-nVidia.JPG
..it looks very normal. The speed is sustained but drops as the blocks get bigger. Average read of 108mb/s. Burst of 236mb/s. VERY normal for a 7200rpm RAID-0.

RocketRaid 2310 (4x Raptor RAID0)..
http://www.wiinter.com/pics/HDTACH-HPT.JPG
..it strikes me as odd that the transfer rate fluctuates greatly, but very consistently, and doesn’t drop off as the blocks get larger. IOMeter corroborates the average speed of 120mb/s and a burst that is LOWER than the nVidea controller.

Or, if you absolutely prefer IOMeter:

Here is a reference test from one of HighPoint’s competitors:
FOUR x Raptor RAID0 - (FastTRAK S150 Controller) - Sequential Read 100%
• 1kb = 60.87mb/s
• 8kb = 158.23mb/s
• 128kb = 214.96mb/s
• 1024kb = 208.35mb/s
http://www.xbitlabs.com/articles/storage/d...s150-tx4_7.html

Here are the results from my 2xMaxtor setup using my built-in SATA RAID0 on the motherboard:
TWO x Maxtor RAID0 - (Asus SATA Controller) - Sequential Read 100%
• 1kb = 42.02mb/s
• 8kb = 79.93mb/s
• 128kb = 83.32mb/s
• 1024kb = 82.61mb/s

And, finally, here are my results with the RocketRAID and the raptor drives:
FOUR x Raptor RAID0 - (RocketRAID Controller) - Sequential Read 100%
• 1kb = 15.81mb/s
• 8kb = 53.12mb/s
• 128kb = 170.34mb/s
• 1024kb = 169.77mb/s

Much slower at the smaller chunks, but picks up a tiny bit on the larger ones. Still doesn’t even come close to the transfer rates of the four Raptors on a FastTRAK S150.

My troubleshooting efforts:

1) There are three PCI-e slots on my Asus P5N32-E SLI, I have now tried this card in two of them (one 16x and one 16x/8x), but it has the same results.
2) I have upgraded the bios of the card to v1.12, and the controller drivers to v1.15; the very latest. Still did not fix the problem.
3) I’ve tried disabling/enabling spread spectrum control, in the mobo bios, for both the SATA and PCI-e. Did not seem to change anything.
4) Using the HighPoint RAID management console, I’ve tried disabling NCQ on the two newer drives that support it. I’ve also tried disabling “Read Ahead” and “Write Cache” in various combinations to see if that helped. It did not!
5) I’ve tried changing the positions and order of drives to different channels (via cables). It was a long shot.

I can’t help but feel the problem is related somehow to the sporadic / fluctuating transfer rates – it just doesn’t seem like it is sustaining transfer. HighPoint support and I went back and forth for a couple months, but they had nothing new to offer in regards to tests.

Thanks for your comments! Those who which to comment about the practicality or safety of RAID0, please withhold those suggestions.

-J

[Sagittaria]

You've maxed out your SATA connection. Anything more than a 3 drive RAID array on a SAT2 bus, is a waste. I found this out with my 4 drive raid array on an IDE cable. 4 drives in RAID-0 performed just the same as a 2 drive setup at about 80mb/s

Notice how your Sustained read is at 120, which is about the maximum sustained transfer over a SATA connection (Yes it says 300 on the box, but thats burst, it cannot sustain that speed, same with USB 2.0, says 480, but you get much lower speeds)

[wiinter]

You've maxed out your SATA connection. Anything more than a 3 drive RAID array on a SAT2 bus, is a waste. I found this out with my 4 drive raid array on an IDE cable. 4 drives in RAID-0 performed just the same as a 2 drive setup at about 80mb/s

Notice how your Sustained read is at 120, which is about the maximum sustained transfer over a SATA connection (Yes it says 300 on the box, but thats burst, it cannot sustain that speed, same with USB 2.0, says 480, but you get much lower speeds)

Ah yes, but then how come the xbit review (and about a half dozen others on google) show a better performance on close to the same setup?

[hardnrg]

maybe you have your stripe size set incorrectly?

[jammin]

maybe you have your stripe size set incorrectly?

That's the first thing I would question as well.
The first time I set up my 2 drive RAID 0 it didn't perform as well as it should have in HD Tach.
I rebuilt the array using a different stripe size and all was well.

This post has been edited by jammin: Apr 3 2007, 04:47 AM

[kingdingeling]

what is a good stripe size? I have 128kb right now...

[hardnrg]

there is no one-size-fits-all... it depends on the number of drives and the sizes of the files you use... basically it's somewhat guesswork, trial and error... but if you set it too big, then you lose out on parallel transfer performance, and if it's too small, you may get performance impacts from the overhead of managing a large number of file segments per file

[wiinter]

Ah yes, but then how come the xbit review (and about a half dozen others on google) show a better performance on close to the same setup?

Can you change the stripe size post-OS install?

[Waco]

Your sustained rate should be a lot higher than 120 MB/s, and your burst should be MUCH higher than what it is as well.

When I had my 3 Hitachi drives in RAID 0 I had over 200 MB/s sustained bandwidth and my burst rate was over 400 MB/s. You should get about quadruple your regular sustained and burst speed with 4 drives.

Then again, even with those higher sustained reads/writes you aren't going to see much of a performance increase...latency is the limiting factor in hard drives today, not bandwidth.

[LivingGhost]

It's not hardware raid...As said in my own thread about raid, you're really not going to get much more performance without hardware raid...

[ClayMeow]

It's not hardware raid...As said in my own thread about raid, you're really not going to get much more performance without hardware raid...

huh? Sounds like it's hardware raid to me. The RocketRaid is a raid controller afaik.

Anyways, as for waco's comments, what drives were you using? Raptors are regular SATA, not SATAII or 3.0GB/s or whatever you want to refer to it as.

PS. When I first saw my RAID graph with the variation, I thought it was odd too. heh

[Nemo]

huh? Sounds like it's hardware raid to me. The RocketRaid is a raid controller afaik.

Anyways, as for waco's comments, what drives were you using? Raptors are regular SATA, not SATAII or 3.0GB/s or whatever you want to refer to it as.

PS. When I first saw my RAID graph with the variation, I thought it was odd too. heh

It's a RAID controller card, however it does not have an on-board processor to handle the RAID processing functions which must be handled by the CPU just like using a SATA controller on your motherboard. This probably doesn't make a huge deal of difference unless you are using RAID level 5 where a lot of parity calculations need to be performed, in which case a RAID controller card with it's own processor will make a huge difference.

I found this tip at Hardware Review for properly setting your stripe and cluster size:

Tips for Improving Performance

Some thought needs to go into planning a RAID array. Small stripe sizes favor large numbers of files being transferred while large sizes favor activities such as video editing etc. Generally a 16K stripe size works best for everyday use. To get the best performance it is necessary to format the Array using a cluster size that is a whole multiple of the stripe size. For example if the stripe size is 16K then it is best to use 16K or 32K as the cluster size. The reason for this is that Windows sends/requests data in blocks that are made up of the cluster size. The RAID controller allocates these to the first free drive in the array in sizes of the stripe size. So a 32K cluster would be split into 2 16K blocks and sent to 2 disks and this would be optimal for 2 or 4 disk Arrays. 4K clusters would have to be accumulated until 16K was ready and then sent to the first disk while the other(s) were waiting for data. In practice this is not too bad as the cache on modern drives compensates for this but the greater the number of drives in an array the greater the need to take such factors into consideration.

Can you change the stripe size post-OS install?

Only if your OS is not on the RAID array. You have to reformat the array to change the stripe size.