hard drives & raid - benchmark and compare!

[Fight Game]

Both ATTO and HD Tach are OS independent. The benchmark results are the same whether the drive is populated or not according to their documentation and the empirical results I have gotten.

 

I know you need an OS to run the programs. What we were talking about is running the program (from the drive or partition with the OS) on another, blank drive (or partition). Once you're running the program, you can choose to run the test on any drive, not just the OS drive (or partition).

 

So you're saying the results will be the same, populated or not. But, from what I've read, you should use a larger stripe/cluster ratio for larger files, so wouldn't you get better results from:

1. Running benchmark with 64/16 on drive with very large files compared to 64/16 on drive with small files, and

2. (vice versa) running benchmark with 16/4 on drive with very small files compared to 16/4 on drive with large files...

 

If this is true, then there's not much since in running the tests on a blank drive. You would have to populate the drive, with whatever you want there, with each combination (128/32, 64/16, 32/8, 16/4) and test several times each to really know. ha! And if you ever change anything, like drive(s), number of drives, raid controller, cpu (maybe?), or the average file size of the data, then all new testing will be needed.

 

I ran a couple benchmarks on each (128/32, 64/16, 32/8, and 16/4) with the OS installed (and didn't have an extra drive to run the OS and program from so was very tedious) and found what we pretty much already knew, that is that the 16/4 is the best bet, for OS, because of all the small files. I thought maybe it would be different since I'm using 4 drives in the array, but it wasn't. I didn't have the patience to install all my favorite games with each scenerio, and test. Besides, I don't know what part of the disk the benchmark is running anyway, so why would I? I guess I need another drive to install an OS and benchmark, and test each scenerio with only my games installed, because this is what I really want to optimise for - loading and playing games. But even if I do this, how would I know if the benchmark is testing an area where there is data, or a blank area.

For this reason, and what you said, I'm thinking maybe the whole idea of using a larger stripe/cluster for larger files, is bogus.

Maybe I should populate an entire array with large games, then I know the benchmark will be testing an area of interest.

[Fight Game]

And you can have different cluster sizes on the array as long as it's partitioned

Ya but then one or more of the partitions won't have the 4:1 ratio.

 

Oh, always load the OS on the test drive with the RAID drives not connected.

Is this because of the fact that the OS likes to write data to all drives during the install? Don't know why this would matter, just for quick testing purposes only.

Or why do you say this?

[soundx98]

Up until two years ago every superfast RAID 0 I saw and everyone told me to use was 16/16. RGone convinced me to try 32/8 as a friend had told him it worked well on Raptors. I would think it would have something to do with the improvements in HDDs over the years and very much because of the newer controllers as well.

 

Seems like every time I use IDE and SATA together when I load OSes something writes where it shouldn't. And that often means a boot issue.

 

I sure hope I am not coming across as a know-it-all, I'm just a newb trying to tell what results I've had testing lots of RAID arrays. And I urge others to find out for themselves by testing their own.

 

As always, these are benchmarks and YMMV

[red930]

To get the very best performance from a RAID-0 setup, the drives should be on separate controllers.

 

The nF4 SATA controller is setup like this...

 

SATA port 1 is primary and SATA port 2 is secondary on one controller.

 

SATA port 3 is primary and SATA port 4 is secondary on one controller.

 

If you are setting up a two drive array, your drives should be on ports 1 and 3 or 2 and 4.

 

According to my testing the results are very close whether you use two drives on the same controller or you separate them.

 

BUT!

 

After running several dozen sessions of each benchmark, I was able to determine that the results derived from using using two controllers gave a reproducible and statistically significant increase in performance.

 

Now here's the number... 2 percent!

 

I know, it's an awfully small increase but it's free.

 

Plus, it doesn't matter which ports you're now using. You can change the ports without having to reinstall the OS. You can switch or even reverse the ports and the array will still function normally.

 

I like keeping my cable clutter to the minimum so having the two array drives plugged into ports 2 and 4 keeps the cables as close to the edge of the board as possible.

 

How's that for benefits. An increase in performance and it looks good too.

 

ExRoadie

[xenolith]

People in this RDX200 thread are saying there's no real world benefits to the SATAII interface. Even over IDE/PATA66... and Rgone is agreeing with them! :confused:

[red930]

The Crossfire board does not support SATA II so theirs no reason to get SATA II drives.

 

People in this RDX200 thread are saying there's no real world benefits to the SATAII interface. Even over IDE/PATA66... and Rgone is agreeing with them! :confused:

[xenolith]

I know, but some are saying (in generalizations) there's no real benefit we would see with SATAII even if it was supported by the Crossfire board.

[red930]

Well, you can take a look at the benchmarks on the Hard Drive benchmark thread and decide for yourself.

 

I know, but some are saying (in generalizations) there's no real benefit we would see with SATAII even if it was supported by the Crossfire board.

[xenolith]

Well, you can take a look at the benchmarks on the Hard Drive benchmark thread and decide for yourself.

 

That's why I ordered four 80GB Hitachi SATAII's

[red930]

That's gonna be a smokin' array on that board. Just remember that you've increased the likelyhood of an array failure by a factor of four with that many drives. Keep your data safe.

 

That's why I ordered four 80GB Hitachi SATAII's

[Fight Game]

Seems like every time I use IDE and SATA together when I load OSes something writes where it shouldn't. And that often means a boot issue.

 

Ya, windows, for some unknown reason, during install, will write data to every drive connected, and if any of those drives are moved/removed, it can and probably will have a boot issue. That's why you should only have the drive(s) connected that you want the OS on, during install. Once the install is complete, then hook upo the other drives. I don't think it matters if it's IDE and SATA together, it just writes to any and all? drives connected during the install regardless.

[Fight Game]

To get the very best performance from a RAID-0 setup, the drives should be on separate controllers.

 

You said this about the SATA, but is this the same for ATA? Would there be a benefit from putting 1 drive in the Primary and 1 drive in the secondary? Right now my 2 ATA drives are on 1 cable going to Primary, and I'm pretty sure the SATA drives are in 2 and 4.