Enter The Matrix: Slice out and get the best part from your hard drives

[Douken]

And a newer Intel southbridge

And at least a pair of these Perpendicular recording drives (or better yet 4 )

The cheapest ones are hovering around $100 each (the 250 and 320Gig models)
http://www.newegg.com/Product/Product.asp?Item=N82E16822148144
http://www.newegg.com/Product/Product.asp?Item=N82E16822148140

$100 each hurts my pocket a lot

[Edit]

ehh? wtf they are the same price the 250 & the 320!!

 

[bing]

bing.. what was your final PCmark05 score.. futuremark is loading VERY slow right now
..edit..
nevermind, I see it didnt give you a final score.

Nope, don't have final score since I ignored the requirement for MPlayer 10 and MS Encoder. Look the result, few failed test.

Brief comparison only HD related result between mine and with the official OcF champion "Spiritedandy" with monster rig AMD 64 X2 3.2 GHz and high end "scuzzy" drives. (at here http://www.ocforums.com/showthread.php?t=458761)


Comparison Result-------------------Mine-----------Spiritedandy (da champion)

HDD - XP Startup(MB/s)-------------13.95-----------12.15 -> No big deal , but still surprising !
File Decryption(MB/s)----------------72.67-----------55.99 -> Depends on CPU,memory and disk, means my disk is hell faster !
File Encryption(MB/s)----------------37.88-----------33.46
HDD - Virus Scan(MB/s)-------------46.94-----------36.41 -> This too !

Of course, my other result like CPU, memory and GFX are sux big time compared to his rig. Anyway, it supposed to be beaten on every aspect in those benchmark right ?

This is software on the SB from what I can tell, the write-back cache option is likely the cause of the CPU hit and that is going to differ with the amount of processing power you have most likely, I'd like to see the scores you could get trying this
If the Raid0 performs this much better with the matrix raid, you might be making up for the CPU hit, wich might be alot less then his depending on the rig a 930 at near 4Ghz shouldnt have much trouble

My 2 cents,

I doubt if this is purely software, I guess it sorta hybrid between "limited" dedicated Raid functionality in the south bridge chip plus good OS integration in the driver software. This "limited" of course can be easily "enhanced" in future generation or once it gained enough popularity !

See there are different features among different Intel chipset family. If it is purely software, they can just enabled all same features for all chipsets without hitting hard on the CPU utilisation. See my prev. result, after OC-ing the FSB, the CPU utlisation is lower from 15% to 10% which is significant IMO.

Strategy wise, maybe this feature is one of the weapon "to differenciate" Intel chipset against other competitor chipset manucfacture such as nVidia, ATI , SIS or VIA.

Of course Intel's end game is targetting AMD, have been browsing other AMD board with "onboard" Raid 0 with Raptor drives ! Still could not find one that beats my Raid 0 burst speed ! Please, don't have the intention to start a flame war here, just could not find it yet.



-

 

[greenmaji]

Your not going to find scores like that, your Raid0 uses only the front end of the drives a regular Raid0 config stands no chance performance wise.

And by saying the software is on the SB, I was saying its a SB raid config, IE hardware raid. Normal software raid is done by the operating system, this is NOT IT! Onboard "hardware" raid's are done though the chipset or a seperate Raid controller chip, this is the southbridge that makes it hardware. Party calculations (for Raid5) and write-back caching would be done by the CPU, a true hardware Raid config would have have its own processing power and ram on the CONTROLLER .
Snugglebear will likely run across this and add details I'm not recalling or even thinking about

Its too bad you didn't install the software to get a score You likely would have done fairly well considering your using onboard video

 

[Xaotic]

Comparison Result-------------------Mine-----------Spiritedandy (da champion)

HDD - XP Startup(MB/s)-------------13.95-----------12.15 -> No big deal , but still surprising !
File Decryption(MB/s)----------------72.67-----------55.99 -> Depends on CPU,memory and disk, means my disk is hell faster !
File Encryption(MB/s)----------------37.88-----------33.46
HDD - Virus Scan(MB/s)-------------46.94-----------36.41 -> This too !

He's using a pair of 36GB Raptors on what is probably a NVRaid controller. Still not bad numbers, but high end SCSI it's unfortunately not...

 

[bing]

He's using a pair of 36GB Raptors on what is probably a NVRaid controller. Still not bad numbers, but high end SCSI it's unfortunately not...

Nope, he is using SiImage SCSI, is this high end SCSI or just "ordinary" SATA Scsi ?

Real "scuzzy" drive is way more expensive than SATA counterpart. I know, sorta more reliable with those MTBF thingee and totally different interface.




See, my lowly humble rig is destined to be beaten at any time "now" and "easily" !

I don't care and I don't even have my PCMark05 score, just want to verify whether I got "real number" and to confirm that my crazy el-cheapo Raid 0 is really showing something, and now it is proven it really delivers quite a "bang" for the "bucks" I paid. Of course sharing it here too, that's it !

Heck, I don't even have a plan to OC my $97 CPU P4 D805 and still happy with the "on board" VGA. hey.. I heard that "YUCK !" word !
You know, my pal used to call it "Grannie Grade" rig ! LOL



-

 

[Xaotic]

Silicon Image does not make SCSI controllers, hence checking Spiritedandy's sig reveals 2x36GB Raptors in his sig. MS is also notorious for listing devices as SCSI when they are attached to a nonIDE controller no matter the interface. PCMark relies on data provided by the host.

I'm very familiar with SCSI, in addition to being in storage development and part of performance teams. Generally, with larger SCSI arrays, I'm concerned with getting 150-200MB/s in STR and good disk access times since I use them for larger files at work. SCSI is not truely about speed, but stability, reliability and data preservation.

The scores are probably mostly accurate, with the exception being Sandra. It has has many issues over the years with RAID implementations and results are distrusted in many places. I've seen benches using high capacity 15K 16MB cache drives in single configuration score lower in Sandra than 7200RPM IDE drives. To do a full profile, IOmeter should probably be run, but it's profiles are very tightly tuned and more based for specific applications and server usage patterns. For general usage, I use ATTO and HDTach to get a general overview of the HDD storage subsystem performance.

 

[bing]

I'm very familiar with SCSI, in addition to being in storage development and part of performance teams. Generally, with larger SCSI arrays, I'm concerned with getting 150-200MB/s in STR and good disk access times since I use them for larger files at work. SCSI is not truely about speed, but stability, reliability and data preservation.

The scores are probably mostly accurate, with the exception being Sandra. It has has many issues over the years with RAID implementations and results are distrusted in many places. I've seen benches using high capacity 15K 16MB cache drives in single configuration score lower in Sandra than 7200RPM IDE drives. To do a full profile, IOmeter should probably be run, but it's profiles are very tightly tuned and more based for specific applications and server usage patterns. For general usage, I use ATTO and HDTach to get a general overview of the HDD storage subsystem performance.

Thanks for the brief insight, nice to know we got an storage guru here !

Could you comment something about my HDTach and ATTO benchmark result ? As you said though for "general usage", simple, is it worth the investment made ?

About IOMeter, yep I tried it too using the script from StorageReview (its complicated with lot things that I'm not quite understand) but I think it is not quite suitable or realistic for profiling a single user at desktop machine performance right ? Or it is ?

Again, if you don't mind to guide this noob to use the IOMeter "properly", I will post the number here.


Cheers
-

 

[Xaotic]

Not quite a guru compared to a couple of people here, but been around several blocks...

Your scores are more than just acceptable, they're excellent. The SATA interface is coming along nicely in terms of performance, and for desktop usage are easily more cost efficient than SCSI or SAS. In terms of investment, it's a good deal. In fact, I couldn't resist a pair of the 320s either. My Yonah board has clearance issues with 64bit PCI slots and 1x PCI-E slots are not sufficiently wide for a decent U320 controller, so it'll be SATA on that box.

IOmeter is good for obtaining specific usage patterns. I use it to test a variety of server patterns for my home servers and for stress testing hardware configurations, but for many people it's overkill since they will never stack IOs as high as the bench is capable of loading.

 

[bing]

-


Thanks, btw those two 7200.10 250GB cost me $85 each !


My envious friend keeps telling me that those crazy result is a "pseudo" performance. LOL


I agree, in "other" scenario, this kind of configuration will be a disaster, by the processing overhead for that 2 volumes in 2 physical drives either in OS, firmware and worst at mechanical level with those poor heads banging around between the two raid regions.


In "my" scenario, again "my" requirement and "my real life" usage , the access frequency at the RAID 1 volume will be minimal since its only for storing/archiving precious personal data such as digicam pics or videocam movies (those are big) plus documents and misc. important files. Yep, I'm lazy to backup to those DVD burner.


And looking at this good price/bytes/performance ratio, I'm plannig to buy another drive to increase the Raid 0 performance "and" to recover the space lost in that Raid 1 by converting it into Raid 5. That 3 drives in total still a lot cheaper than single Raptor drive.


Now noob question again, any idea what to expect by adding another drive into the two drives Raid 0 ? Remember, I am and will be using the "front" speedy area of each disk for this Raid 0. In term of avg.seek, Raptor still beats Seagate, but mine already 9 ms, will adding more drive will help the avg.seek score ?


To others if you got 3 new identical drives and using Intel matrix raid capable chip, please if you got a chance to test matrix raid (0+5) and post the result.


-

 

[greenmaji]

Now noob question again, any idea what to expect by adding another drive into the two drives Raid 0 ? Remember, I'm will be using the "front" speedy area of each disk for this Raid 0. In term of avg.seek, Raptor still beats Seagate, but mine already 9 ms, will adding more drive will help the avg.seek score ?

It would be a guess, do it, bench it, post it
Good luck

 

[bing]

It would be a guess, do it, bench it, post it
Good luck

LOL, yeah but I'm broke after I spent it for LCD 21" screen.

Anyway, found a good reading here : Raid 0 Guide

One thing interesting about RAID 0 from above page is this :

Random Read Performance: Very good; better if using larger stripe sizes if the controller supports independent reads to different disks in the array.

Random Write Performance: Very good; again, best if using a larger stripe size and a controller supporting independent writes.

Ok, currently I'm using the largest strip size supported by Intel Raid which is 128K and all those 7200.10 are NCQ enabled. My best guess is the NCQ is utilized by Intel driver, as they claimed in their web site. LOL

Now, look at other 7200.10 owner's performance in Raid 0 (using whole drive) which almost all above 11 ms, while mine is constantly at 9ms (narrow & front speedy area), isn't this result looks promising ?

Yeah, guessing as you said, I'm "guessing" that by adding more drive while squeezing tighter at front disk area will improve the seek time.

Hopefully beats the Raptor's seek time by this cheap drives array. Time will tell or until I have enuf money ! LOL


-

 

[bing]

-

Interesting finding, check it out here :

Seagate Barracuda 7200.10 SATA Product Manual (PDF) 1.6MB



Quote from page 29 :




To storage guru, any idea with those Recording Density differences among 7200.10 drives, do they matter ? Is it significant ?

Mine are 250GB, say compared with 300GB or 320GB ?


-

 

[greenmaji]

Higher is better, you likely have half a 500gig in platers.
Your 250gig looks like it would be slightly faster then the 320Gig (they must have used mulitple smaller platers to get the 320Gig capsity.)
Its just one spec. that can change things for the better or worse.

 

[Xaotic]

As above and geometry changes are a double edged sword. Firmware optimizations and other factors typically have larger effects. As with most varieties of engineering, there are tradeoffs everywhere.

 

[bing]

Updates !

Now, here are proofs of my idea on gaining good performance from the disk by using only the fastest part of the disk to serve this purpose.

Result on the Raid 0 at two Seagate 7200.10 250GB drives.



Note: This are with 16K strip size, as you can see the CPU utilization is quite high, I lost the one with 128K strip size captured screen, it was 5%
After lot of testings at different strip sizes, seems like for this "particular" Intel ICH Raid, the best still at 128K same as other reports from many sources in the web.
Yes, I'm aware of common practice for 16K strip size but for this setup 128K is the best.


Now, this is HDTune result same as above Raid configuration but with few markings that I made for explanation.



About those markings :

- Looks like the fastest part for transfer rate still about 10% of the front area 147.7 MB/s and starting to degrade to the sh*ttiest part at the rear (71.0 MB/s). Pointed by that 2 blue arrows.

- The yellow box marking is the targetted area that I will make it into Raid 0 drive, decided on 20GB (approx. 4% from total size) for nice round number. If your disk is got different size like 320GB you can make based on how big & acceptable performance you'll need, and make the decision based on HDTune result like above.

- Again, in that yellow box area, we can expect the average transfer rate will be hovering bout 150 to 140MB/s and look at those scaterred dots in the box, the average seek times will be around 7 to 9 ms.




The result by using only 20GB front area for the Raid 0 volume :

As expected, HDTune result looks exactly like the "zoomed" yellow area above.




HDTach result




To other that want to try this Intel Raid, just make sure you enable the write back cache at the Matrix Console, just right click the Raid 0 volume as this pic shows :




Conclusion :

- Using this Matrix Raid if you're equipped with two big hardisks or more, 3 ideal, 4 drives is the best if you can afford it . It is possible to gain a very good performance while you'll get spare/left over space for Raid 1 (2 disks) or Raid 5 (3 or more disk) for protection. Heck, even this kinda setup will serve the "other" purpose such as getting a really good benchies number ! ..hint..hint...

- So far, as I could find through the web, it can be beaten only by that expensive iRAM solidtate disk or a well configured high end dedicated Raid card. Of course with the similiar performance class of harddisk & system. If you use with latest Raptor, the result will be really crazy. But that is different story and with different "funding" level too.

- The slowest part of the computer still at the harddisks, this is one of the area and alternative (cheap one IMO) that shows it can be improved significantly by using this method.

- Finally , just a reminder, this kind of Matrix Raid setup is NOT for everybody. Make sure it fits into your requirement & real life usage.


If you got 2 new fresh hard disks and with an Intel Matrix Raid capable mobo, please try it and post your result here, just for comparison sake. Also I'm very curious & eager to know how is the new generation mobo like 975X coupled with Conroe will perform in this Matrix Raid ? Again, please !


-

 

[bing]

Some additional references I found :

Guide for creating the Intel Matrix Raid
ExtremeTech Report: Making Matrix RAID Work (November 30, 2004)

Look at the date, geez.. I wish I read this before, went thru the pains from the scratch, learning by doing by myself !


This are the benchies report of Intel onboard Raid ASUS P5WDG2-WS with Pentium 955 EE VS 3 dedicated SATA RAID controller cards from Adaptec, LSI Logic, and Promise
Just read the interesting summary.

ExtremeTech Report: Workstation RAID Card Shootout (June 15, 2006)

-

 

[M Diddy]

I'm assuming this performance increase in the RAID 0 is due to the partitioning of the drives and making the RAID 0 on the fastest seeking part of the drive???

I'm currently running a RAID 0 on my 2 WD 160GB SATA 2's, but just that RAID 0, it's not partitioned at all with the RAID 1. I got a great BURST speed, much better then yours actually Bing, but am wondering if something's not right....

 

[M Diddy]

Just for info.... Here's my 32MB run wtih HD tach on my RAID 0....

These are on 2 160GB WD Sata 2's with 8 MB cache and a 128KB Stripe size....

 

[bing]

I'm assuming this performance increase in the RAID 0 is due to the partitioning of the drives and making the RAID 0 on the fastest seeking part of the drive???

Yes, sort of but it is not exactly like "partitioning" as ordinary in OS level, since it is handled by that Intel driver and tightly integrated into the south bridge chip that makes it efficient.

At your HDTach result, the fastest part is at front about 130MBbps and degrades gradually to 75MBps at the end of the drive. Since you use the whole drive, the the average seek is mediocre at 12.9ms since they span througout the drive.

The idea of using Matrix Raid is to "chop" out the best part and form RAID 0 and use them as boot, page file and program for better responsive rig. Check my previous post, managed to squezee out the seek time from 13 ms to 8ms, not bad eh ? Also my OS boot time is almost 10 seconds !

Its sort of permanently isolate & keep them in the fastest region instead to let them slowly scattered throughout the drive which will degrade the responsiveness. Remember, that pesky Windows Update is constantly updating our system files, and it juggles and scatters those tightly packed system files since they're installed at the 1st time.

The effect is obvious, our Windoze is getting slower from time to time.

Not the holy grail though, this method has a disadvantages since you will get a smaller and larger slow drive rather than single contigous drive like your current one.

Agree, a "bit" complicated and as I mentioned throughout this thread, it is not suitable for everyone. Its up to you whether it fits in !

I'm currently running a RAID 0 on my 2 WD 160GB SATA 2's, but just that RAID 0, it's not partitioned at all with the RAID 1. I got a great BURST speed, much better then yours actually Bing, but am wondering if something's not right....

Finally, it shows up here with that "spooky" burst !

Nah, my rig is nuthing and it is expected to be beaten ...he..he..

It is normal, that is the typical Intel ICH RAID signature on the huge burst that even beats a dedicated PCI-E RAID card like Areca 1230 !

 

[M Diddy]

Well, I decided to give it a shot.... Not that complicated for me as I deal with Mid Range servers on a daily basis and know my fair share about RAID.

I set up a 50GB RAID 0 and the other 125 to RAID one.... I'll let the pic speak for it's self.....