6 x 1.5GB + Perc 5i + RAID-5 = Fast Storage

[Stereodude]

Gear:

• 6x Samsung Spinpoint F2EG HD154UI 1.5TB 5400 RPM 32MB cache
• Dell Perc 5/i PCI-E x8 SAS RAID controller with 256MB cache & battery
• 2x 32pin SAS to 4 SATA cables

The array is a single RAID-5 volume with a 512k stripe.

Single drive benchmarks for reference point:

ATTO:

HD Tune Read:

HD Tune Write:

6x1.5TB RAID-5 Array benchmarks:

ATTO:

HD Tune Read:

HD Tune Write:

All benchmarks were run in their default configuration. Also keep in mind that HD Tune only tests the first 2GB of a drive.

Not a bad setup if I say so myself. I will do more benchmarks later after I finish stress testing it with IOmeter.

 

[Mercutio]

Those Perc5 controllers really are a bargain, but if you're using the hardware RAID function, you probably should buy a second one. If that controller fails your data is hosed.

 

[Stereodude]

The same is true of any RAID card though I'll admit I never thought about it that way.

 

[Mercutio]

The same is true of any RAID card though I'll admit I never thought about it that way.

That is why I continue to use software RAID even on cards that can do it in hardware. I like my data to be recoverable if necessary.

 

[Stereodude]

So, apparently ATTO is largely worthless. You can run it 5 times in a row and get drastically different results each time.

Now I'm seeing if I can set up IOmeter to basically do the same thing.

 

[Stereodude]

I may have spoke too soon. It seems the RAID card was busy doing things in the background which might have been causing the results to bounce all over.

Two successive runs on my 3ware RAID card produced very similar (though terrible) results.

 

[Stereodude]

Some 3ware ATTO results at different queue depths:

 

[ddrueding]

Looks like a poor RAID-5 implementation to me. My 9650SE-16 does much better writes.

 

[Stereodude]

Looks like a poor RAID-5 implementation to me. My 9650SE-16 does much better writes.

It's not exactly what 3ware advertised or claimed for the 7506 series.

 

[ddrueding]

I think I have an 8506 I could compare it to when I get back...

 

[Stereodude]

I think I have an 8506 I could compare it to when I get back...

Thanks for the offer, but it doesn't really matter since I'm retiring the card this weekend. My new array is up and running having passed stress tests, lots of benchmarking, and all that jazz. I'm copying all the data to it now. Here's the final benchmark. It's an Iometer run set up to mimic ATTO but with a much larger data length to take the 256MB cache on the card out of play.

 

[ddrueding]

That is very healthy performance, Stereodude.

 

[Stereodude]

Well, the Perc 5/i has been replaced by a Perc 6/i. :erm:

I've also ordered two more 1.5TB 5400RPM Samsung drives. I will end up running 8 x 1.5GB + Perc 6/i + RAID-6. :bomb:

I just don't know when to stop. :rotfl: However, without a port multiplier (which costs a lot more than the used Perc 6/i) I'm limited to 8 drives per card, so I guess I've hit the limit on this project.

 

[Stereodude]

Here's the ATTO style benchmark (from IOmeter) of my 8 x 1.5TB RAID-6 array on a Perc 6/i. It's slower than the 6 x 1.5TB RAID-5 array on a Perc 5/i with .5K, 1K, and 2K reads, but is faster everywhere else (look at the numbers, not the length of the bars).

8 x 1.5TB RAID-6 array (Perc 6/i):

6 x 1.5TB RAID-5 array (Perc 5/i):

I never tested the 6 x 1.5TB RAID-5 array on the Perc 6/i, so I'm not sure if the performance loss is from the RAID level, or the RAID card.