P2000 G3 FC LFF - Raid Question

[Patrick Neuner]

Hello,

I have a Question about the RAID Configuration of our P2000.

Given Situation:

P2000 with dual controllers
8 x 2TB SATA Drives (Hitachi HUA72202)
connected to
C7000 Blade Center with 6 Blades

predicted workload: Web-Sites, Mail-Storage

The OS, DBs and other write intensive Stuff is located at our EVA4100.

The Question:

We need to set the correct RAID Level for our needs. (unfortunately you can't change Raid-Level later with the P2000) But we are not sure about the Performance we get.

Of course Raid 10 would be interessting - but we don't want to loose 50% capacity.

We thought about the following Configurations:

1) Raid 5 with 7 HDDs and 1 spare
2) Raid 6 with 7 HDDs and 1 spare
3) Raid 50 with 2 x 3 HDDs and 2 spare

We would prefer Raid 50 because of the better random write performance over Raid 5 or 6. But unfortunately the HDD configurations we can do with Raid 50 is not desireable (always lot's of spares OR no spare as 12 HDDs is max with LFF HDDs) - or would Raid 5/6 perform better because there are more HDDs involved?

Which configuration would perform best of those 3?

I hope you can help me with this Question.

Thank you, bye from Tyrol (Austria)
Andreas

[Uwe Zessin]

Performance? If you want performance, then don't select large, slow disk drives !

With that large disk drives I would select RAID-6 - if the P2000 allows it, in a 8-disk configuration. If all bays are taken, buy an extra, external "spare-disk". Otherwise there is a high risk that during recovery from a disk failure an unreadable-error happens on another disk - which results in data loss.

[Patrick Neuner]

@UWE: speaking about best performance possible with this configuration ... no Question that higher performance is achievable with 15k FC/SAS HDDs. But for the "real" High Performance Stuff we use our EVA4100 (equipped with 15k FC SFF HDDs) ... but I guess everyone want's to get the best out of his stuff, right?

Any ideas how long a rebuild process of the Raid 5/50 would take?

Data safety while rebuild is the point why we also consider Raid 6.

[Patrick Neuner]

Only thing what drives me away from Raid 6 is that I read lot's about the VERY BAD random write performance - and Web-Site Traffic / Mail Traffic would only be small, random traffic.

But I personally don't have that much experience with MSA SANs nor with RAIDs including that large HDDs.

[Patrick Terlisten]

Hello Andreas,

I would also recommend a RAID 6, but without a dedicated spare. It depends on you support level. If you have a 4h onsite, I wouldn't use a dedicated spare.

RAID 6 is the worst case for small random writes. So I hope that you mail application doesn't do small random write IOs (like Exchange 200x).

Regards,
Patrick

[Patrick Neuner]

Hello,

now we conducted a few Benchmarks and are a little irritated.

Can it be true that a RAID 1 (2 drives) is more than 2x faster than a RAID 6 with 8 drives?
If this is true we would go for some RAID 1 Disks as the rebuild time for a failed RAID 1 with 2 TB should also be a lot faster than a RAID 6 rebuild with 8 drives ... ?

Also there were only marginal differences between RAID 5 and RAID 6 - irritates me a little as I thought that RAID 6 is a lot slower in terms of writing?!?

Thank you
Andreas

[GundW]

Can it be true that a RAID 1 (2 drives) is more than 2x faster than a RAID 6 with 8 drives?

Yes. There is no parity calculation required for RAID1, the data is simply written to both disks. With RAID6, there are two parity stripes, so each write to to the logical disk requires a parity calculation and this parity to be written to each other disk in the array. So yes, writes are slow. Reads should be quick, because you have 8 heads working at it.

If this is true we would go for some RAID 1 Disks as the rebuild time for a failed RAID 1 with 2 TB should also be a lot faster than a RAID 6 rebuild with 8 drives ... ?

Depends what you want. If you want fast reads and you have worked out that 8 drives does not saturate the bus, then RAID6 is for you. Also, RAID6 can withstand two disks failing before the array is lost. RAID1 can only take 1 disk failing before things get dark. This is well explained in an article on Wikipedia with performance considerations, although not a very technical article, it's a good start.

Also there were only marginal differences between RAID 5 and RAID 6 - irritates me a little as I thought that RAID 6 is a lot slower in terms of writing?!?

As above. RAID5 can only take one disk dying before it get's dark, consider RAID5+1 hot spare to be equal to RAID6. Except with RAID5, writes are quicker because there is only one parity stripe.

The other consideration is that RAID5 yields the most usable disk. You already know that RAID5 will give you 12TB, RAID6 10TB and RAID 50 a whole lot less at 4TB.