hard drives & raid - benchmark and compare!

[soundx98]

@FG, shouldn't be any difference in a single RAID 0 array or two on the controller.

You will take a slight hit when copying from one to the other of course.

 

just set up the RAptor array 1st and back it bootable, and then set up another array with whatever other two drives. Piece of cake.

[acurax]

Alright so I've read through parts of this thread and tried to search through it for stripe/cluster sizes but I'm still not sure.

 

I'm gonna be setting up a SATA RAID 0 array on the NVRAID controller next week sometime with two Seagate 7200.10 320GB drives. Pretty much the drives are going to be used for PVR recordings, DVD/media storage, some video editing. My OS and games will remain on the 74gb Raptor.

 

Since this is my first RAID setup, I have some questions:

 

1.) What should my stripe / cluster sizes be?

From what I've read a 4:1 ratio is recommended for benches (i.e. 16k stripe / 4k cluster). Since I'm not running an OS, and am dealing mainly with large file transfers across the array, should this be different? Does the 4:1 ratio still apply?

 

2.) Does it matter which SATA ports I use? (e.g. can I throw them on 2 & 3?) My Raptor is on SATA1 right now.

 

System specs are in sig...thanks for any suggestions.

[red930]

Here's the best advice you will ever get.

 

Build and test until you find what works for your particular setup.

 

This is how we came up with our recommendations for the nF4 boards and arrays listed in this thread.

 

I just got my hands on a pair of Seagate 320GB 16MB SATA 3Gbps drives with the new Perpendicular recording technology.

 

I've got them installed as a secondary RAID-0 array in Aileen so I can rebuild the stripe and cluster sizes. If I find anything that's different from our prior observations, I'll build an OS array with them and test again to validate.

 

At this time all I can say is that they are damn fast.

[acurax]

Alright I can do that. Just have to get some 'real-world' benchmarks.

 

Thanks, and I guess I'm still wondering if it matters which SATA port they go on. All the SATA ports are their own 'channel' so to speak correct? or are 1&2 and 3&4 connected somehow?

[red930]

All the SATA ports are their own 'channel' so to speak correct? or are 1&2 and 3&4 connected somehow?

There are two SATA controllers on the nVidia nF4 chipset. Ports 1&2 are separate channels on one controller. Ports 3&4 are separate channels on the other controller.

 

I haven't been able to detect a difference between an array of two drives on the same controller or across the two controllers using benchmarks or application testing.

 

This shows the controller and channels have more bandwidth available than current hard drives can utilize.

[Guest Kobalt]

You can say that again.

 

Today, I was updating my 160GB drive to a 300GB HD, and just for the heck of it, I wanted to copy all files. This took almost 2 hours at a blazing average speed of 26MB/sec.

This was from a Seagate SATA 1.5 (on SATA port 1) to a seagate SATA 3 (on SATA port 3).

Then I wanted to see if it was a 1:1 copy, so I did some verifying, and that went a bit faster as expected, but still slow!

 

So that is some real world #'s, not some crummy benchmark.

 

We just need to get some HDs that can read/write at over 100MB/sec, and I will be happy for 6 months.

[NEOAethyr]

Hey I hope you guys don't mind me coming in here and chatting a tiny bit...

I don't have a raid setup, nor plan on getting one.

But..., I'm interested in collecting good hd benching programs.

 

I'm also interesting in getting those new seagate 7200.10's...

In the future though.

 

Anyways I dl'ed the free ver of hdtach right, and I got some odd results.

Can you guys check this? :

http://www.savefile.com/files/8371150

 

The sustained read scores seem a tad high, for these drives I think.

Not a huge deal, but the cpu utilization of 0%?

I don't think that's right...

 

Plus my scores are hiher then ref scores in that program, which seems kinda odd because normally ref scores in a program are at best.

Mine on the other hand, well, look at the pics, I was dl'ing torrents, dual priming...

 

I did'nt mod anything to do with the pci bus or hd controller, latency or grant wise, or anything like that at all.

Using the onboard ide controller on the nfii lpb, with sw ide's ver 667.

 

Both ports filled up, the 2 wd's on the same controller, the segate with my dvdrw.

 

All of my drives are filled up mostly, one with around 30gigs free, the other around 20gigs.

The seagate has around 20gigs free.

 

I was'nt at high speeds or anythuing, 250mhz fsb, slack timings, 2.1ghz cpu.

You get the idea, nothing special, slow actually.

 

I'm just wondering if I should try a diff program.

 

 

Edit:

Hmm, hmmm..

 

That 1600jb of mine, I wonder if that's why it has a hard time being detected sometimes after I clear my cmos.

Maybe...

 

I have all my drives on there side right now.

The 1200 and the seagate are setup where the arm swings up to the center.

The 1600 the arm would swing down to the center.

I wonder if I reverese that if I can fix that sucker...

And bring it's ramdom access time back down.

 

I should bench it the oppiset way, then upside down and upside right...

I'm puting together another case for myself, snapped a dremal bit so I have to wait still, but if the drives prefer being setup in one direction I'll have to go with that...

 

I will check that out later on today or tonight or something, when I got time anyways.

I gotta do tons of cleaning in my house today :.

[red930]

Today, I was updating my 160GB drive to a 300GB HD, and just for the heck of it, I wanted to copy all files. This took almost 2 hours at a blazing average speed of 26MB/sec.

That's not a terribly slow transfer speed when you take all the variables into account.

 

Copying the data could result in reading that data multiple times before it was actually transferred and verified.

 

Were the source and destination drives properly defragmented? Are the file systems on the drives at 100%? Did you have any background apps running? Were the Antivirus and Malware applications scanning each file?

 

The simple fact that the OS will verify each file copied will cut your throughput.

[Guest Kobalt]

That's not a terribly slow transfer speed when you take all the variables into account.

 

Copying the data could result in reading that data multiple times before it was actually transferred and verified.

 

Were the source and destination drives properly defragmented? Are the file systems on the drives at 100%? Did you have any background apps running? Were the Antivirus and Malware applications scanning each file?

 

The simple fact that the OS will verify each file copied will cut your throughput.

It is terribly slow when your waiting for it to get done.

 

The only thing running was the standard windows crap (needed system files), and explorer. I killed everything else. No indexing service, No antiviurs, no firewall, no browsers, no nothing that would affect the results.

 

Oh, windows doesn't verify files on a copy or move operation. At least, I think I read that someplace.

[red930]

It is terribly slow when your waiting for it to get done.

 

The only thing running was the standard windows crap (needed system files), and explorer. I killed everything else. No indexing service, No antiviurs, no firewall, no browsers, no nothing that would affect the results.

 

Oh, windows doesn't verify files on a copy or move operation. At least, I think I read that someplace.

Besides the CRC check done, the OS monitors the transfer of data to verify completion of the transfer.

 

This is why you find the data rate lowered on certain drives if the OS detects transfer errors.

 

An Optical drive might start at Ultra DMA-2 Ultra33 but if there are data errors it will fall back to a lower rate until it stops getting errors.

 

A couple of months back I had a fairly new HP rig hit the shop with "constant drive activity". After verifying that there were no nasties on the rig it turned out the data cable was bad and the drive was running in PIO mode. A new cable fixed it right up for the standard 1 Hour Bench Fee!

[NEOAethyr]

To add onto what ExRoadie said, windows also monitors I/O.

In many ways.

 

1st bing the main I/O counters for disk operations.

The 2nd being the performance counters for the hardrive controller.

 

The crc32's, like ExRoadie said.

 

Last access times...

File creation dates...

Truncated extentions...

Long file names and times stampes for FAT...

8.3 ver's of the file names for NTFS...

 

NTFS USN journals...

Disk performance counters...(yes many of these, many many).

 

I'm sure there's much more.

Not even including the workload over IRQL's when using the ones related to drive access.

 

 

ExRoadie, actually windows will continue to fall back all the way to pio mode, the errors will not stop .

Every single time I've ever seen that issue happen, it's not because of the transfer mode(Well in your case sorta, but it still fell back all the way to pio did'nt it? ).

So hence, it will fall back all the way eventually.

Once you hit pio mode, all heck breaks lose, since some controllers absolutely hate pio mode.

 

That and the lower the transfer mode you go, the lower the peak bandwith, and the more cpu utilization used to transfer files.

 

 

Oh yeah indexing..

Yep.

There's tons of that, the service, the indexing logs.., etc.

Even then it's still there.

 

 

Kobalt

2hrs to copy 160 gigs..

Wow man, that's along time...

I'm not bragging or anything but it takes me at tops 40mins to copy around 120gigs, normally around 20-40mins, depends on how much I copy.

I dn the exacts, but 2hrs is very very long, it's never taken me that long to do a straight copy.

 

I mean it may take me all night to copy stuff to here and there, remake partitions, etc etc, to get it all setup.

Then once it's all setup, copying the files to there correct location does'nt take to long.

 

For example a 80gig partiton to copy to a 100gig partitoin only takes me around 20mins probably.

Another 20mins to do my other drive's big partition.

Ruffly anyways, maybe add another 10-20mins, I forgot.

Not an hour though.

 

I just did this about a month ago...

 

It takes much longer on this seagate, it's a really old drive though.

 

 

To tell you the truth.

It's probably the I/O size setup.

To little and you get more cpu utilization.

Less on small teeny tiny files.

But truthfully, look at an avg game and see how many huge files there are...

Even in windows, though most files are small, small enough where 32k cluster would be optimal, but in real world performance, in my case anyways, 64k straight up wins in all cases.

 

Cluster size.

MBS/Block mode.

Stripe size.

 

In any case you have it all said and done so whatever .

[green-man]

hi raid noob here,

was after a bit of advice on sum sort of raid setup,ive been looking through this thread. but thats only thrown up more ?? for me. i was thinking about a 3 disc raid 5 setup in my head,as remember seeing raid5 in the dfi manual for my board. i was after better disc read speeds but was'nt to worryed about write speed and wanted a bit of sercurity.so raid 5 would have been perfect. but after reading the first copple of pages of this threads it made me think raid 5 is different to i remember. thought it was like data is written to two disc's simultaneously so it'll only be the speed of a single disc or a little bit slower but u get sum security as two disc with same data. when data is read though its simular to raid 0 supports block sizes, so you get sum extra speed? naw ive got my 3 samsung 250gb and ive seen the sil3114 on my 3200dr is slow with raid5 n e way wether im right or not about how it works.

so naw im not sure what to do?.should i go for 2x disc in raid 0 on the uli1575 and the other as back up?. or 3x raid 0 setup as sum of u guys seem to be liking?, seems sweet speed wise but i dont understand what going on there? could sum1 please explain for me? or do i go buy a raid pci/pci-e card for raid5 ? cheers:)