3 X Raptor 150 in Matrix Raid 0 - Results and Information

[dominick32]

Well, I must say I am happy with my results but we have hit a new conclusion and discovery with Intels Matrix Raid 0. We all assumed that with increasing the capacity by adding more drives, while at the same time keeping a SMALL boot slice in Raid 0, the matrix would continue to decrease access time and increase sustained read. For the Raptors I went from an 8.5ms access time in single drive configuration to a 6.2ms Matrix Raid 0 access time. We all though by adding another drive we would theoretically continue to decrease access time. This assumption is false. My access time remained exactly the same and for this benchmark, even increased .1ms.

My older Raid 0 Matrix benchmarks had my 2 X Raptor 150's HDTached at:
175.5 Sustained Read and 6.2ms access time.

My newer Raid 0 Matrix benchmarks of 3 X Raptor 150's are HDTached at:
260 Sustained Read and 6.3ms access time.

So, that equates to an increase of 85 MB/s transfer and a mild increase (not decrease) of .1 ms access time. I must say, I am happy but I truly believed like everyone else that access time would drop below 6.0.

Also, anyone new to Matrix Raid I would highly recommend reading Bings thread here:http://www.ocforums.com/showthread.php?t=467848

To anyone interested in Matrix raid, I recently answered a Private Message question that another OCForums member asked me. Here was my response:

THE MATRIX
Its called Matrix Raid and its quite awesome actually. It lets you physically create slices (not partitions) out of your drives which forces the system to read/write from the fastest portion of the platters. In a sense faking creation of a physical drive smaller than it actually is. You really have to try it out for yourself to actually look and decipher the difference between a slice and a partition. A slice would not allow the outer portions past the drives max capacity to be read in HDTach or other benchmarks where as a partition would still technically include the entire hard disks capacitys.

In my case, I am using 3X 150 gig Raptor X's with a total capacity of 450 GB in Raid 0. I will be setting up my rig with 2 slices, which tells the system they are 2 seperate HDD's.

Slice #1 = 20 GB Total (6.6 GB's each drive) Raid 0
Slice #2 = 430 GB Total (143.3 GB's each drive) Raid 0

Since Slice #1 is only using the first 6.6GB's of each drive to create a 20GB boot drive, each platter gives you incredible sustained read/write transfer rates and should theoretically drop random access time significantly for me with the speedy Raptors. (Please note: Random Access time stock is 8.5ms on the Raptors) It puts the old NVRaid AMD Raid Controller to shame.

As far as real world performance, the gain is very noticeable. With the introduction of Matrix raid and HDD slicing it is even more apparent.

Slice #2 implements the rest of the 430GB Capacity into a humongous Raid 0 disk with a very good STR and access time, but no where near as fast as the first boot slice. I will be using the 2nd 430GB slice for data/software/music/etc.

Our super speedy Slice#1 will be used for windows xp boot, benchmarking, gaming, and heavily accessed main applications.

Wow... Suprisingly my drives keep up quiet well.

I hit 232MB/s and have seeks of ~9.5ms Not bad for 3x 320gigs with a 150Gig Partition hehe.

Its the access time that I am really after, not so much the STR. A access time of 6.3ms is very quick and definitely the first thing you notice with the raptor raid setup. First you start out with a single Raptor 150 with an average RAT between 8.1 and 8.5. Putting that into a 2,3, or 4 drive matrix config cuts the seek time down completely by 2ms. That is very substantial. Heck anything at or below 9.5ms is noticeably faster than the 11,12,13,14 ms HDD's. But, your setup is killer. Especially for the pricetag and cutting down the seek to 9.5ms is a definite noticeable difference in performance. I have almost come to the conclusion that when using Raptors, unless you are after video editing or server level I/O's, a 2 X 150 Raptor setup may be the best config in the high end price/performance category versus more of them. It seems that anything after 2 drives actually may increase access time, when our theorys think otherwise. I am a gamer and benchmarker so I am fully after dropping access time. Those STR's dont really mean that much to me although its nice to have close to 300 MB/s on 3 drives. I would think a 3 or 4 drive Raptor Matrix setup would be absolutely killer for a hardcore video editor. For a hardcore gamer, 2 drives is looking like the hot ticket.

I may even downgrade to 2 drives again because I was hoping for a sub 6ms access time. And I dont think purchasing another 150 will help my config, or my wallet. lol

Wahoo quoted Hehe

Well isn't it suppose to increase times when you add raid just due to the controllers latencey and communicating back and forth?

Yes, but..... lol
And the but is the cool part. With the Matrix and the ability to continue slicing down drives it should decrease, at least by theory. Example:

Single Raptor(non raid) - 1 Drive, 150GB single drive partition = 8.1 to 8.5ms
2 X Raptor 150 - 2 Drives, 10GB each drive, 1 slice = 6.2ms
3 X Raptor 150 - 3 Drives, 6.6GB each drive, 1 slice = 6.3ms

Our theory was that once we actually reach the third drive and continuously decrease the used capacity of each drive that the readable portion of each platter would shrink, and continue to decrease access time because the drive would have less of a distance to tinker on the platter. As you can see in a 2 drive configuration, that theory holds true. But it looks as though after 2 drives on the matrix, we have something else impacting seek times as you stated such as controller, and latency, or perhaps the actual lowest physical seek time on the disk. If that makes any sense. Technically by looking at that data you would think that 6.6GB readable portion per drive would seek better than 10GB per drive on the platter. Maybe anything 10GB and under produces exactly the same results? I will need to do some more research. I was truly hoping to prove myself wrong and get those killer sub 6ms access times.
Cheers.

 

[krag]

Very nice! I can't wait! But i'm gonna have to wait until ol kents becomes available.

i am extremely impressed with Intel's new hardware. Your raid benchies are phenominal!

 

[tuskenraider]

(Please note: Random Access time stock is 8.5ms on the Raptors) It puts the old NVRaid AMD Raid Controller to shame.

I get 7.8 with a single and 8.0 on a pair in RAID0 on a VIA chip...........

 

[dominick32]

I get 7.8 with a single and 8.0 on a pair in RAID0 on a VIA chip...........

You have either 36 or 74g raptors. I have a Raptor 150...
Maybe I should have clarified what I meant by "stock". Stock meaning, non Matrix raid 0. A regular Raid 0 array and partitioning setup. ie: NVRaid, VIA, etc.

 

[dominick32]

Very nice! I can't wait! But i'm gonna have to wait until ol kents becomes available.

i am extremely impressed with Intel's new hardware. Your raid benchies are phenominal!

Thanks for the response Krag. I cant wait to finally see what your setup is going to be. Your nickname should be TOTL.... "Top of the Line".

Kentsfield coupled with Dual 8800GTX's? What else is on your plate? hehe

 

[deathman20]

Wow... Suprisingly my drives keep up quiet well.

I hit 232MB/s and have seeks of ~9.5ms Not bad for 3x 320gigs with a 150Gig Partition hehe.

 

[bing]

Wow, great result Dom ! Any plan to add up the fourth ?

I remember you mentioned testing out those strip sizes, any conclusion ? And also could you please take an Atto snapshot at 32M of that slice 1 of raid 0 ?

 

[dominick32]

Testing out stripe sizes was a stalemate on this Intel setup. I initially tried 128K to duplicate my dual drive setup. Only installed the operating system and HDTach. It produced identical results to my 64K setup. I guess on Matrix setups you really will only notice the stripe size on server level I/O's...

Here is the 64K Atto shot bing:

 

[dominick32]

Wow... Suprisingly my drives keep up quiet well.

I hit 232MB/s and have seeks of ~9.5ms Not bad for 3x 320gigs with a 150Gig Partition hehe.

Its the access time that I am really after, not so much the STR. A random access of 6.3ms is very quick and definitely the first thing you notice with the raptor raid setup. First you start out with a single Raptor 150 with an average RAT between 8.1 and 8.5. Putting that into a 2,3, or 4 drive matrix config cuts the seek time down completely by 2ms. That is very substantial. Heck anything at or below 9.5ms is noticeably faster than the 11,12,13,14 ms HDD's. But, your setup is killer. Especially for the pricetag and cutting down the seek to 9.5ms is a definite noticeable difference in performance. I have almost come to the conclusion that when using Raptors, unless you are after video editing or server level I/O's, a 2 X 150 Raptor setup may be the best config in the high end price/performance category versus more of them. It seems that anything after 2 drives actually may increase the access time, when our theorys think otherwise. I am a gamer and benchmarker so I am fully after dropping access time. Those STR's dont really mean that much to me although its nice to have close to 300 MB/s on 3 drives. I would think a 3 or 4 drive Raptor Matrix setup would be absolutely killer for a hardcore video editor. For a hardcore gamer, 2 drives is looking like the hot ticket.

I may even downgrade to 2 drives again because I was hoping for a sub 6ms access time. And I dont think purchasing another 150 will help my config, or my wallet. lol

 

[deathman20]

Wahoo quoted Hehe

Well isn't it suppose to increase times when you add raid just due to the controllers latencey and communicating back and forth?

 

[dominick32]

Wahoo quoted Hehe

Well isn't it suppose to increase times when you add raid just due to the controllers latencey and communicating back and forth?

Yes, but..... lol
And the but is the cool part. With the Matrix and the ability to continue slicing down drives it should decrease, at least by theory. Example:

Single Raptor(non raid) - 1 Drive, 150GB single drive partition = 8.1 to 8.5ms
2 X Raptor 150 - 2 Drives, 10GB each drive, 1 slice = 6.2ms
3 X Raptor 150 - 3 Drives, 6.6GB each drive, 1 slice = 6.3ms

Our theory was that once we actually reach the third drive and continuously decrease the used capacity of each drive that the readable portion of each platter would shrink, and continue to decrease access time because the drive would have less of a distance to tinker on the platter. As you can see in a 2 drive configuration, that theory holds true. But it looks as though after 2 drives on the matrix, we have something else impacting seek times as you stated such as controller, and latency, or perhaps the actual lowest physical seek time on the disk. If that makes any sense. Technically by looking at that data you would think that 6.6GB readable portion per drive would seek better than 10GB per drive on the platter. Maybe anything 10GB and under produces exactly the same results? I will need to do some more research. I was truly hoping to prove myself wrong and get those killer sub 6ms seek times.

 

[tuskenraider]

Let me throw a little something into the mix................Maybe your newest Raptor isn't performing as it should be and slowing down the rest? I mention this because I got a new 36GB Raptor a couple months ago and it benched pretty good(71.4MB/s STR and 7.9-8.0ms seeks). I got a second for RAID and benched it and it was faster in STR(78.6) and seeks(7.7-8) consistantly. Small difference, I know, but I paid for the best, I should get it on both drives. So now I considered my first one as not performing to spec and RMA'd it to WD last week after I told them that was not acceptable for this drive. Replacement drive is 76.5 STR and 7.8 seeks consistantly, which I find acceptable. In RAID0 I went from 143.8 STR to 149.1 and 8.3-8.4 to 8.0-1 in seeks. So maybe you want to test your drives individually and see if you have a clunker.

 

[dominick32]

Let me throw a little something into the mix................Maybe your newest Raptor isn't performing as it should be and slowing down the rest? I mention this because I got a new 36GB Raptor a couple months ago and it benched pretty good(71.4MB/s STR and 7.9-8.0ms seeks). I got a second for RAID and benched it and it was faster in STR(78.6) and seeks(7.7-8) consistantly. Small difference, I know, but I paid for the best, I should get it on both drives. So now I considered my first one as not performing to spec and RMA'd it to WD last week after I told them that was not acceptable for this drive. Replacement drive is 76.5 STR and 7.8 seeks consistantly, which I find acceptable. In RAID0 I went from 143.8 STR to 149.1 and 8.3-8.4 to 8.0-1 in seeks. So maybe you want to test your drives individually and see if you have a clunker.

I guess anythings possible. But I highly doubt that is the case. If you look carefully. A single Raptor 150 comes from the manufacturer with a sustained read of 85 MB/s. If you look at my array, before adding the third drive to the mix I was at 175.5 MB/s. By adding the third drive my STR increased exactly 85 MB/s to 261 MB/s. That could tell me that the sustained read is exactly where it should be and the drive is not defective.

I would also expect the array to experience severe Raid lag problems and a much higher access time if there was indeed problems with my drive.

 

[tuskenraider]

I know it's unlikely based on the data, but the possibility was/is there as seek times don't affect STR. Might be worth a check if you get bored one day.

 

[dominick32]

I know it's unlikely based on the data, but the possibility was/is there as seek times don't affect STR.

Very true but I would have to assume if the drive was faulty that STR would correlate to a depleted seek time as well just like it did when you had to RMA yours. Dont you think?

Anyway, I appreciate the feedback.

 

[tuskenraider]

No problem and definately, but I'm sure you've seen stranger things in this hobby.

 

[emboss]

Two things:

First, in your case you do not need matrix raid. You could achieve exactly the same result using a single RAID0 array and partitioning. Matrix RAID is only useful if you want different RAID levels on the two volumes.

Second, the reason the access time increases is that there's two components to the access time: the seek time and the rotational latency. The rotational latency is increasing as you add drives (assuming that the HDTach seek tests is reading a sufficiently big block to hit all the drives, which it appears to do). One drive has 3ms average rotational latency, and it increases to 4ms for two drives and 4.5ms for three drives.

So the seek time components for the arrays are 2.2ms for the two-drive array, and 1.8ms for the 3-drive array (give or take 0.1ms or so). So the seek time has decreased, as expected, but has not decreased enough to counter the increased rotational latency.

 

[Max0r]

HARD DISK ENVY

 

[krag]

Two things:

First, in your case you do not need matrix raid. You could achieve exactly the same result using a single RAID0 array and partitioning. Matrix RAID is only useful if you want different RAID levels on the two volumes.

Second, the reason the access time increases is that there's two components to the access time: the seek time and the rotational latency. The rotational latency is increasing as you add drives (assuming that the HDTach seek tests is reading a sufficiently big block to hit all the drives, which it appears to do). One drive has 3ms average rotational latency, and it increases to 4ms for two drives and 4.5ms for three drives.

So the seek time components for the arrays are 2.2ms for the two-drive array, and 1.8ms for the 3-drive array (give or take 0.1ms or so). So the seek time has decreased, as expected, but has not decreased enough to counter the increased rotational latency.

Can you show me where regular Raid can achieve 350Mbs on read & write? I was/am unaware that regular raid can do that well.

 

[bing]

Yeah I'm also curious, cause in my case tested intensively, the pure software based Raid 0 like in Windows Server version couldn't beat the "plain Raid 0" using ICHxR based Raid.