Matrix Raid - Oh Yeah

Finished the last bit of my upgrade to c2d with 2 more hitachi sata2 80gb.

Giving:-
4x80gb Hitach1 Sata2
Using Intel Matrix Raid
2 Raid stripes :-
1st "Fast" Raid using 1st 17gb of each drive = 64gb in Raid0 128kb stripe size std cluster
2nd "slow" Raid using remaining space = 242gb in Raid0 128kb stripe size std cluster.

Fast Raid
1) 8M Burst=3,336 Read=203 Access=9.7 CPU%=5
http://homepages.nildram.co.uk/~chsm...mbFastRaid.JPG
2) 32M Burst=3,456 Read=237 Access=9.7 CPU%=4
http://homepages.nildram.co.uk/~chsm...2mFastraid.JPG

Slow Raid:cool:
1) 8M Burst=3,475 Read=176 Access=12.1 CPU%=4
http://homepages.nildram.co.uk/~chsm...bSlow260gb.JPG
2) 32M Burst=3,461 Read=186 Access=12.2 CPU%=6
http://homepages.nildram.co.uk/~chsm...2mSlowRaid.JPG

Think the hitachi & Matrix is a pretty decent combiination.
For more on Matrix Raid see Enter The Matrix: Slice out and get the best part from your hard drives
& Intel® Matrix Storage Technology Only real down side is the matrix raid has to be configured within XP, so i have a small xp partition on 1 of my data drives.

Interesting post. Just setup my first raid. Have one volume setup as raid 0 total 300 gb (2x150 raptors).

Have used HD Tach and my burst speeds read about 250 mb/s.

For the fun of it I right clicked on the volume and enabled write back cache and it jumped to 3000 mb/s. kin hell

The graph was well choppy though. Just briefly checking out that thread, I'm guessing that you need to setup more than one volume, with the first being below a 100 gbs. Is that right?

Cheers for the info.

RLM

There is, speed-wise, no difference between a single RAID volume partitioned, and two RAID volumes. The access times are only superficial.
It's a nice feature, but that's all it is - a feature.

I didn't know it has to be configured from XP, I configured mine from the BIOS (though I have the ESB2 Server ICH, not ICH7/8R, but I doubt there is any difference).

@RLM

Quote:

---

I'm guessing that you need to setup more than one volume, with the first being below a 100 gbs. Is that right?

---

Yep it's easier to use a % of your total real estate(GBS) Thie above sizes were based on my 1st attempt where i went for a 85gb fast raid
http://homepages.nildram.co.uk/~chsm...ol0_single.JPG
You can see the distinct drop at the end around 77.5% mark. so 77.5% x 85Gb = 66GB. Resizing to this has the benefit of lifting the benchmark perrformance of both drives as the 2nd raid is getting faster access space.

So my overall factor% (the X from Bing's chart) = 20.6% 65/(4x80)

@alfaunits - Don't think you've got the point m8. That's why i added the links at end of 1st post.

see
Bing's Chartfrom 1st link above
http://homepages.nildram.co.uk/~chsm...id/Matrix3.PNG
As u can see the"Fast Raid" can give a far faster raid than average as it 's using the fastest parts of the component drives. I'm using this partition for my main os & swap file.
Hdtune shiows min / max & avge so there is a reasonable consistenvy of high access speed.
Fast Raid HDTune
http://homepages.nildram.co.uk/~chsm...e0-69gbx2c.png
The 2bd raid volume which i'm using as raid0 atm usese the ramainder & is slower. Here there is a far bigger range of speeds with of course a corresponding lower average.
http://homepages.nildram.co.uk/~chsm...1-250gbx2C.png

The point is the matrix raid allows up to 2 raid volumes to be stripped across a collection of the same drives as shown in the chart.

luck:D

These numbers just frankly don't make sense. I am referring mainly to the burst speed here.

Why? Well, lets assume you are using 4x full-speed SATA II ports here, for a total bit rate of 12Gbps, or 1.5GBps. Yet your burst speeds are showing in excess of 3GBps! That's over *twice* the maximum, theoretical bandwidth (which you will never hit).

Clearly there's something else going on here. You said you set this up in Windows... I'm almost wondering if the burst speed is coming from caching data to RAM, because it's *not* coming from your hard drives.

With that said, the rest also comes under suspicion...

I'm surprised you didn't notice that your burst numbers - being more than 10x what you should see with one drive - were at least somewhat suspicious.

There are a lot of people ignorant about Matrix RAID over at OCForums and they love their benchmarks over there. I gave up trying to explain how 15GB of data on a regular or Matrix RAID array will have the same performance. Instead of using a Matrix array or a partition for that matter, they should invest in a good defragmenter.

I'm also using a 69GB slice/partition w/ 2 74GB ADFD's. It def feels snappier than when the whole drive's used. As for the crazy burst rate, I thought the SATA limits were per port? And o&o did a wonderful job w/ my single raptor, but had a helluva time w/ a mirrored array. Perfect disk threw a fit too. Anyone successfully used Diskkeeper's full version on their arrays?

I accept the burstrate looks a bit silly but it apears common in 2 versions of software, it isn't that impartant imo.

There are some odd burst rates scattered around have looked thru the thread here lawrywild - Here's my 2x Seagate 7200.8 200GB

Accept that some of it is statistical averaging. If i set up a 65gn partition on a clean single raid0 & put my primary os on it then i would expect to see a similar performane to my "fast Raid" stripe.

But i like the matrix flexibility in being able to have a 2nd stripe which may be raid1, 5 etc., At the moment i have my less used os's on it xp64, Vista 32 & 64 rc2 + a data partition. So when running SP2 am getting the max benefit

But like Jodiuh says

Quote:

---

using a 69GB slice/partition w/ 2 74GB ADFD's. It def feels snappier than when the whole drive's used

---

is an absolute fact.

I'd love to see some timed actual proof, which not one person has done that I've seen. Considering it's the same chipset, whether Matrix or not, I'd bet there'll be not one difference. Install an OS and programs on an array, time them, image it to a data drive, reimage to a new array and time again. If I had an Intel ICHxR chipset, I'd do it in a heartbeat.

Quote:

---

Originally Posted by Jodiuh

I'm also using a 69GB slice/partition w/ 2 74GB ADFD's. It def feels snappier than when the whole drive's used. As for the crazy burst rate, I thought the SATA limits were per port? And o&o did a wonderful job w/ my single raptor, but had a helluva time w/ a mirrored array. Perfect disk threw a fit too. Anyone successfully used Diskkeeper's full version on their arrays?

---

The SATA limits are per port, 12Gbps = 4x 3Gbps ports.

I have used Diskeeper successfully on a mirrored array myself (though if you're using matrix raid, it could be the controller is having some issue with that I suppose, however strange that would be).

@Supershanks:
The burst rate isn't "a little silly", it's downright "horribly wrong".

Quote:

---

I'd love to see some timed actual proof, which not one person has done that I've seen. Considering it's the same chipset, whether Matrix or not, I'd bet there'll be not one difference. Install an OS and programs on an array, time them, image it to a data drive, reimage to a new array and time again. If I had an Intel ICHxR chipset, I'd do it in a heartbeat.

---

Sorry you mean what exactly??

Quote:

---

The burst rate isn't "a little silly", it's downright "horribly wrong".

---

:) :D U should know what us brits are like 4 understatement;)

I dont think the burst numbers are real, it does this on the ICH7/8 as well. I was running something similiar on P5W/Commando and performance sure doesnt feel like those burst speed. As for the partition slice, it seems the program is just reading off average throughout the partition. So if you have a shorter partition, the beginning will be faster and your average will be higher just before it starts to dip. So this slice method is just manipulating the benchmark or is it real world performance? I havent done this method so I can't comment. The only real data I did was a full partition of ICH8R w/ 4x raptors, i installed vista with a stopwatch in about ~14:56 min by the time desktop shows. With the areca setup and same drives, about ~11:20 min.

http://i55.photobucket.com/albums/g1...ptor_raid2.jpg

That burst is just some caching going on. Someone on this forum looked at this a while back when people were posting insane bursts with their areca cards. It's not really there.

Quote:

---

Originally Posted by Supershanks

Sorry you mean what exactly??

---

That showing those benchmarks has nothing to do with actual performance, that perpetuating an uproven theory that Matrix RAID actually increases performance without objective proof is sorta irresponsible? A partition can do the exact same thing but I don't see people claiming how fast their system improved when they've done that? Why, because HD Tach can't benchmark partitions.

@Speederlander

Quote:

---

That burst is just some caching going on. Someone on this forum looked at this a while back when people were posting insane bursts with their areca cards. It's not really there

---

Makes sense all caching is on, normal running, except av/fwall down & network disabled while testing.

Reading round i have seen it said that

Quote:

---

intel from time to time do tricks with their drivers giving wrong info..

---

Just coming from AMD rig i've no idea if there's any validity in this comment??

@syne_24

Quote:

---

s for the partition slice, it seems the program is just reading off average throughout the partition. So if you have a shorter partition, the beginning will be faster and your average will be higher just before it starts to dip.

---

Totally Agree - Which is where i got my original slice size from

&

Quote:

---

Post#8
If i set up a 65gn partition on a clean single raid0 & put my primary os on it then i would expect to see a similar performane to my "fast Raid" stripe.

---

What i see as a benefit of matrix is the abilility to have a 2nd raid slice with a different Raid 1,5 tec., to get the best options whilst gaining the benefit of the max number of drives.

@tuskenraider-

Quote:

---

That showing those benchmarks has nothing to do with actual performance, that perpetuating an uproven theory that Matrix RAID actually increases performance without objective proof is sorta irresponsible? A partition can do the exact same thing but I don't see people claiming how fast their system improved when they've done that? Why, because HD Tach can't benchmark partitions

---

I aren't making any political or propaganda statement, nor am i a believer in any conspiracy theories:D . I'm just a relative newbie to using 1) c2d - new build & 2) Using raid + matrix raid.
Just saying my hd performance feels faster:woot: don't convey much does it?? Thought others may like to share & the software shots where a reasonable way 2 do it:)

Quote:

---

Finished the last bit of my upgrade to c2d with 2 more hitachi sata2 80gb

---

1st line of this thread

Cheers:D

Quote:

---

Originally Posted by Supershanks

@tuskenraider-
I aren't making any political or propaganda statement, nor am i a believer in any conspiracy theories:D . I'm just a relative newbie to using 1) c2d - new build & 2) Using raid + matrix raid.
Just saying my hd performance feels faster:woot: don't convey much does it?? Thought others may like to share & the software shots where a reasonable way 2 do it:)

Cheers:D

---

I'm certainly not here to belittle the joy of your system, and you certainly have the grasp of what Matrix RAID was intended for. It is just that I've seen people take those benchmarks and ignore all the reasoning and logic behind the results and make lots of unproven claims. I would think truth and objective results serves everyone best. With that said, enjoy your new setup.

Quote:

---

With that said, enjoy your new setup

---

Cheers m8:D

Quote:

---

I've seen people take those benchmarks and ignore all the reasoning and logic behind the results and make lots of unproven claims

---

Wasn't aware of this b4, understand where your coming from

Quote:

---

I would think truth and objective results serves everyone best

---

Is the only way we properly learn, which is what I intend doing & 1 of the primary reasons i'm here:D

luck:D

Behold the DRAMA of RAID! :soap: :thumbsup:

My understanding is that the main benefit of using a small partition on the outer cyclinders of the disks (which can be done on any single disk or any RAID controller, not just Intel's Matrix) is that you effectively gain very good seek times on that partition because when limited to that partition, the heads have to move very little to cover the utilized surface. All goes to hell as soon as you utilize the other larger partition because then your heads are back to seeking across the full platter.

Thus if you want super fast seek times, you should define a partition for your system that is as small as possible and ONLY use that. It's very expensive to ignore the rest of your disk's storage capacity this way, but it will get you wicked seek times.

It should be noted that this trick will not improve STR... it may make your HD Tach curve look relatively flat but that's because you are looking at a very small subset of it.

Supershanks... what is the rated seek time for those Hitachi drives? I'm assuming around 12ms? Notice how your fast partition has seek times around 9.7ms? That (as I understand it) is the only benefit your realize from doing this. And to shave those 2.4ms off the seek times (which is a 20% reduction in latency), you have to ignore 80% of your disk capacity! Worth it?

very nice raid setup.. :)

I envy you.. :( DFI uses slow promise card

@erwinz

Quote:

---

very nice raid setup.. I envy you.. DFI uses slow promise card

---

Thnx m8, i very nearly got that board. Hope everything else is ok.


I switched to a single raid to prove

Quote:

---

Post#8
If i set up a 65gn partition on a clean single raid0 & put my primary os on it then i would expect to see a similar performane to my "fast Raid" stripe.

---

http://homepages.nildram.co.uk/~chsm...ingleRaid1.JPG Follows exactly as expected
the rest of the single raid speed curve follows theat of the 2nd slice
http://homepages.nildram.co.uk/~chsm...2mSlowRaid.JPG

The burst rates are seriously effected as Speederlander said

Quote:

---

That burst is just some caching going on.

---

So I then tried with Write back cache disabled Comparison
http://homepages.nildram.co.uk/~chsm...Cache_Comp.JPG

About time someone posted a graph comparing the two and performance is unchanged, just as expected. :)

Notice his STR changes quite a bit w/ the cache on tho. I'll finish up the Windows install, throw some games on there, and time the loading to see what effect cache has there. Perhaps I'll do some unzipping too.

Quote:

---

Originally Posted by Serra

I have used Diskeeper successfully on a mirrored array myself...

---

Thanks, I'll give that a go.

What i can't quite get my head round is the no cache Bottom Pic.
I would have expected it to be an almost constant -x mb/sec slower than the cached performance but the uncached runs pretty constant ~160ish until it then merges with cached performance just b4 279gb.

Any offers??:D

I don't understand why write back cache affects STR at all. Isn't write back cache designed to write to the disk when it's idle while the HD-Tach benchmark is a READ test? I suspect something wacky is going on with HD-Tach.

Quote:

---

Originally Posted by tuskenraider

About time someone posted a graph comparing the two and performance is unchanged, just as expected. :)

---

The transfer rate is unaffected, but the access times on the smaller (faster) partition are signficantly improved because the heads need to traverse less of the platter to access data.

Quote:

---

Originally Posted by Serra

These numbers just frankly don't make sense. I am referring mainly to the burst speed here.

Why? Well, lets assume you are using 4x full-speed SATA II ports here, for a total bit rate of 12Gbps, or 1.5GBps. Yet your burst speeds are showing in excess of 3GBps! That's over *twice* the maximum, theoretical bandwidth (which you will never hit).

Clearly there's something else going on here. You said you set this up in Windows... I'm almost wondering if the burst speed is coming from caching data to RAM, because it's *not* coming from your hard drives.

With that said, the rest also comes under suspicion...

I'm surprised you didn't notice that your burst numbers - being more than 10x what you should see with one drive - were at least somewhat suspicious.

---

His numbers are bogus and this has been discussed before here. The above mentioned benchmarks do not reflect real life performance due to different read patterns etc, and matrix raid seems to really throw things off in them. Really hd benchmarks are useless for the most part. Test on real applications as thats the only valid result data.

Quote:

---

Originally Posted by virtualrain

The transfer rate is unaffected, but the access times on the smaller (faster) partition are signficantly improved because the heads need to traverse less of the platter to access data.

---

Well for benchmarking yes, but if you put the same amount of data on either setup and at least defragment them, both setups will have about the same access speed to get to said data. 15GB of data on a 20GB or 200GB partition is still 15GB. Yes, if you didn't defragment the drive, the data will be more scattered across the platter, but who is a PC geek that doesn't do such a thing, let alone have a better third party defragmenter.

@epion2985

Quote:

---

His numbers are bogus and this has been discussed before here

---

u mean i just invented them or are u talking about the burst rates, which we have already shown is down to the write back cache.

Matrix RAID uses RAM as cache so that's from where high burst speeds come.

Only speed benefit from Matrix RAID1 (vs. normal RAID1) comes from effective shortstroking which lowers random access time.

Quote:

---

Originally Posted by Supershanks

What i see as a benefit of matrix is the abilility to have a 2nd raid slice with a different Raid 1,5 tec., to get the best options whilst gaining the benefit of the max number of drives.

---

I think you wouldn't want to use software (maybe better word would be "non-hardware") RAID5 for anything else than read...
http://techreport.com/reviews/2005q4...d/index.x?pg=7


As for synthetic benchmarks... those are useless for knowing real performance:
Theoretically RAID5 increases write speed much over single disk (because of striping) but in reality even hardware controllers struggle to rise real write speed above single disk and in case of smaller files it's actually slower.

Interesting linksw EsaT thanks:)

Quote:

---

Originally Posted by EsaT

Only speed benefit ... comes from effective shortstroking

---

If I could only convince my girlfriend of this! :D :hump: (sorry, couldn't resist!)

Thanks!...i installed the Intel Matrix Storage Manager (software),enabled write back cache and got a speed bump!
See specs for RAID setup, i didn't create two RAID partitions as per the nature of this thread...i wanted to see if my normal RAID array would benefit from enabling write back cache, and it did.

I went from 140MB/s to 152MB/s average read speed with write back enabled. I don't have two RAID arrays setup...a single one, but i have partioned the single RAID array during Vista setup. C drive has 120GB while D has 18.5GB.

Here's the HD Tach from an Intel Bad Axe 2 (ICH7R) using half the total size of 2 ADFD 74 Giggers w/ a 64K strip size and write back cache enabled.

http://img.photobucket.com/albums/v2...uh/2788078.jpg

So you're only using 69.2GB of the full 138.5GB?
I guess the increased speed bump must be you're using the fastest part of the HDD's.

Yeah, I use the other half in a mirror for stuff.

would it be possible with 2x 36 gbs raptors hd? I would like to make a first partition fast ( raid0) to boot up the OS and the rest of the raid , another RAID0 with all the rest of data.
I also have a seagate 320 gbs for backups and data, what should i do?

Quote:

---

Originally Posted by Aceit

would it be possible with 2x 36 gbs raptors hd? I would like to make a first partition fast ( raid0) to boot up the OS and the rest of the raid , another RAID0 with all the rest of data.

---

Yes, if you have a motherboard with an intel southbridge capable of Matrix raid (ICH7R or ICH8R).

Quote:

---

Originally Posted by Aceit

I also have a seagate 320 gbs for backups and data, what should i do?

---

I would make the first partition for swap and temporaries, then all the rest for data ad backups.