8 x 1.5TB + Perc 6i + RAID-6 = Faster Storage

[Stereodude]

Gear:

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

The array is a single RAID-6 volume with a 512k size stripe.

8 x 1.5TB Perc 6/i RAID-6 Array IOmeter benchmark (ATTO imitation):

For comparison here are my previous / other RAID arrays. Note the scale is different for all the graphs.

6 x 1.5TB Perc 5/i RAID-5 Array (512k stripe) IOmeter benchmark (ATTO imitation):


5 x 500GB 3ware 7506-8 RAID-5 Array (64k stripe) IOmeter benchmark (ATTO imitation):

 

[ddrueding]

Those are remarkably good write speeds considering it is RAID-6.

I'm in the middle of creating a RAID-10 array with 6 of those drives on an Areca controller with 2GB cache. It will be interesting to compare the results.

 

[ddrueding]

Verifying a RAID10 array? Annoying....

 

[Handruin]

Have you tried any performance tests if you pull/fail the two parity drives? What are you planning to use the array for?

ps those numbers look very good!

 

[Stereodude]

It will be interesting to compare the results.

Do you want my "Imitation Atto" IOmeter .icf file to test with?

 

[Stereodude]

Have you tried any performance tests if you pull/fail the two parity drives?

No, I didn't do that. That would be an interesting test, though I probably won't do it.

What are you planning to use the array for?

Just file storage nothing special or needing the speed.

 

[Handruin]

Does that give you about 9TB of usable space on a single volume? Is this related to the GigE tests you were also running in a different thread? I guess I'm picturing you keeping this machine in a basement and store files via network?

 

[ddrueding]

Do you want my "Imitation Atto" IOmeter .icf file to test with?

That would be great. I had issues setting up IOMeter the last time I tried using it. When this array has finally finished initializing (2 more days?), I'll give it a shot.

 

[Handruin]

Why does it take so long to initialize a 6-drive RAID 10 ??

 

[Stereodude]

Why does it take so long to initialize a 6-drive RAID 10 ??

I don't think the RAID level matters. It has to write (and read?) all sectors of all the discs. That takes the same amount of time regardless of the RAID level.

 

[ddrueding]

I think SD is right, though most of my arrays are AID0, so I don't have this issue.

Edit: Thinking about it, most of my arrays are AD0, since they are SSDs

 

[Stereodude]

When this array has finally finished initializing (2 more days?), I'll give it a shot.

2 days!?!?!?

I did a full foreground initialization my 8 x 1.5TB RAID-6 array in just over 5 hours. It started 17:54:43 and finished at 23:01:50 on the same day. That's 5 hours, 7 minutes, and 7 seconds.

 

[Stereodude]

I think SD is right, though most of my arrays are AID0, so I don't have this issue.

Why's that? You don't have to initialize AID-0?

 

[Stereodude]

That would be great. I had issues setting up IOMeter the last time I tried using it.

IOmeter wants a blank un-partitioned disk. It will show up blue in the Disk Target list (not yellow with a red line through it) Other than that, it's pretty straight forward (I think).

Here's a link to a zip of the ICF file.

You'll need to check your target drive (since it won't find mine in your system) and you might want to move the slider on the results display from infinite to 5 seconds if you want to see the results as it progresses. Leave the ramp time (30s) and run time alone (5m) to ensure an apples to apples comparison. You save the .CSV from the run and then can pull out the MBps numbers out of the .CSV when the run is complete.

 

[ddrueding]

Which parts of this data should I be looking at for a comparison?

http://spreadsheets.google.com/pub?key=tisPRbqRTASBNlnlZSvqg8g&output=html

 

[Stereodude]

The MBps column.

I put it into my template for you. link

By the way, your results aren't too impressive. :bstd:

 

[ddrueding]

Yeah, there might be some settings I should change. I haven't done anything to it, really.

 

[ddrueding]

Oh, and one of the 1.5TB Samsungs failed on its second power-up, so it switched over to the hot spare already

 

[Stereodude]

Here's your graph for your RAID-10 array:

 

[ddrueding]

Why's that? You don't have to initialize AID-0?

Sorry, didn't see this before. No, you don't. It warns you that any data on the drives will be destroyed, and then the array is live. There is no redundancy, so there is nothing to verify.

 

[ddrueding]

Here's your graph for your RAID-10 array:

Thanks Stereodude. That is pretty lame. Perhaps I'll switch it over to RAID-6 and see what it's capable of.

 

[Stereodude]

Sorry, didn't see this before. No, you don't. It warns you that any data on the drives will be destroyed, and then the array is live. There is no redundancy, so there is nothing to verify.

Oh... I would have assume it would have at least made sure it could write to all the sectors on the drives.

 

[ddrueding]

Oh... I would have assume it would have at least made sure it could write to all the sectors on the drives.

Not in my experience, but I have never tried to create a RAID-0 on an expensive controller, just the onboard stuff.

 

[ddrueding]

Initialization time in RAID-6 looks to be a third of RAID-10...looks promising.

 

[Stereodude]

Thanks Stereodude. That is pretty lame.

What was your stripe size?

 

[ddrueding]

64K

 

[Stereodude]

64K

That's kind of small, which might account for the poor results. Could you set one stripe size for the RAID-1, and another for the RAID-0?

 

[ddrueding]

Possibly. I haven't looked at the manual yet, just using the "quick create" mode.

 

[time]

The RAID10 write performance seems too low to me. Is Writeback disabled or something?

Not sure how useful some of these tests are for database work. AFAIK, it's worth keeping an eye on the 8kB block performance. With this in mind, the RAID 6 alternative may not offer as much as at first glance.

Cool benchmarks though, Stereodude.

 

[Stereodude]

Not sure how useful some of these tests are for database work. AFAIK, it's worth keeping an eye on the 8kB block performance. With this in mind, the RAID 6 alternative may not offer as much as at first glance.

Cool benchmarks though, Stereodude.

There are lots of other IOmeter patterns better for showing database performance. This is intended to be an ATTO clone showing 100% Sequential reads and writes at different block sizes. It's similar to CrystalMark Disk also, but much more comprehensive.

 

[ddrueding]

Another day and a half for that array rebuild

 

[Stereodude]

You really need to get a Perc 6/i and play with it.

 

[ddrueding]

I've never had such a hard time finding the pricing and options for something I know the name of on Dell's site. Still looking...

 

[P5-133XL]

I just went to Dell-> small/medium business-> electronic accessories. Once there I entered perc 6/i in the search text box and found it immediately.

Dell Perc 6/i

 

[Stereodude]

I've never had such a hard time finding the pricing and options for something I know the name of on Dell's site. Still looking...

You can buy working pulls for a lot less via e-bay or the buy/sell section of some forums. Mine was $110 shipped with a battery & cable (but missing the PCI bracket).

 

[ddrueding]

Now setting up an 8-drive RAID6 with a hot-spare. Can someone here give guidance regarding the "Greater Two TB Volume Support" setting when creating the array? The options are "No", "64bit LBA", and "4K Block". I believe "64bit LBA" means support for GPT and is the one I want?

 

[Stereodude]

I think you want 64bit LBA. I'm not sure. I had no such options on my Perc 6/i.

 

[ddrueding]

Sounds good. That is what I went with and it is now at 2.8% finished initializing. I went with the maximum (128k) stripe size.

 

[Stereodude]

So, what are the results?

 

[ddrueding]

That I completely forgot about the array? I'll see what I can do when I get back to the office.