SATA Storage in RAID10

First ensure that you're operating fully cached.  Like all cats are gray at midnight, all drives are the same behind the cache.

What capacity are the drives?  If they are 1 TB or larger, RAID 6 is how 'high availability' is achieved.  If they are lower capacity changing the RAID level will likely not have an effect on availability.

What size are the RAID groups?  If they are large RAID groups, such as (8+2)s, you have the option of making (4+2)s which will have a greater likelihood of a full-stripe write.  That will save you a few I/Os.

Finally, you could short stroke the drives.  If they're not using the capacity, they won't miss losing a lot to get the performance.

Hope this helps.

Ok, guy's and girls, it's a pair of HDS AMS1000's, so not Clariion... But I think it's a generic question.

> What capacity are the drives?  If they are 1 TB or larger RAID 6 is how 'high availability is achieved.  If they are lower capacity changing the RAID level will likely not have an effect on availability.

750GB

> What size are the RAID groups?  If they are large RAID groups, such as (8+2)s, you have the option of making (4+2)s which will have a greater likelihood of a full-stripe write.  That will save you a few I/Os.

Even larger: 16+2 !!!! We were thinking about creating either 8+2 or 8+8. the thing is that at the moment we have a drive failure and the performance hit is enormous. The rebuild is already running 9 days and it's only at 22%, but I must admit the priority is set to (extremely) low.

> Finally, you could short stroke the drives.  If they're not using the capacity, they won't miss losing a lot to get the performance.

The thing is that performance is an issue. ATM we're using about half the capacity == short stroking, but rebuilds are taking way to long and we need more IOps. Especially write performance is very bad, but that was to be expected with RAID6.

Just wanted to know if anyone has ever encountered RAID10 on SATA. We need to reconfigure anyway and we have the option to move to RAID10, which would greatly enhance rebuild times plus write penalty will drop from 6 to 2. Any objections ?

As a general comment - SATA are fine IF you do not expect great performance - if you set the expectation that SATA will not perform as well as fiber channel, then you should be OK. If the application is really performance demanding, then you will be disappointed - SATA have slower seek times, spin rate, etc.

These aren't any faster than what you get in a PC - 7200 RPM. Putting a random workload on these disks will just highlight how slow they really are.

My 2 cents

glen

Hi,

there is no issue in using sata in raid 10.

Its just not often seen in the wild as sata are generally chosen because they are the cheaper option and raid 10 is the more expensive option when it comes to raid levels.

But no technical issues with using it.

My customer bought way too much SATAN drives, so now he's stuck with bad performance and half of the RAID Groups are still empty, so working from what we now have and trying to make the best of it, I want to make optimal use of the drives, so that could be RAID10.

Hi, as already been stated performance on SATA is never going to be that great, they are intended more for capacity than performance. Ok, that was the obvious, but if you have a 16+2 RD6 group, and you want to improve WR performance, I would split this into a 2x(4+4) RD10 and stripe LVM them, or whatever its called in the HDS world. If this was a Clariion you could also go with METALuns, but again I'm not sure if this option is available on your HDS array. On any 'large' RG, rebuild times is always going to be an issue, obviously more so on RD5/6 than RD10.

Note: IOP per spindle is calculated from:

- Average (rotational) latency

- Average read seek (latency)

- Average write seek (latency)

- Rotational Speed

nothing in these parameters mentions TYPE, so the reduction in performance on SATA is not due to the interface type but mainly that SATA are limited to 7200rpm. They are also limited to 3Gb/s, as opposed to 4Gb/s on 15K fibre drives.

It all depends on your workload patterns. Changing a 16+2 RD6 to a 8+8 RD10 would definetly reduce the total number of IOPs available from the RG. The total (back-end) IOPs available depends on number of spindles; again stating the obvious here.

Lots of papers have been written on the advantages/disadvantages of RD5, RD6 and RD10 and comparing and contrasting each. If you read several of them you'll end up even more confused since they mostly tend to contradict each other.

Based on my own past experience, I found good data alignment, striping across multiple RGs, back-end loops and 4/8 data spindles (i.e. not counting parity spindles) to give very good performance.

Good luck !

Your story is clear, but also obvious. This is all known stuff.

Smaller RG's are good for performance, but indeed IOps per RG are less, but when you create META's this is solved. IOps per GB is larger in small RG's compared to large RG's !!

And about rebuild times: I'm looking at a rebuild in such a (16+2) SATA RAID6 RG and it's at 32% and it took the machine 2 weeks to get there.....