Loudenj
2 Iron

Raid 3 for backup

Hi All,

Recently we had an external consultant on site review the new RG\Lun lay out for out incomming arrays (cx4-240's). They suggested a number of things that on the surface seemed ok, untill i saw their recommedation for the Backup to Disk RG. Currently we have commvault as our backup solution with a 9 drive raid 3 on 1TB Drives. I set this up based on the best prac. for commvault on a Clariion  (emc document). The new solution suggested is a Raid 6 RG. When I question this I was told due to the random read\write nature of backups raid 6 is better. I was under the impression that any backup would be sequental write to the backup drive, and a restore would be be a similar read.

Is there something I have missed? Any comments would be most appericated.

Cheers

JL

Labels (1)
0 Kudos
12 Replies
RRR
6 Indium

Re: Raid 3 for backup

RAID3 was widely used back in the "old" days indeed, because of the sequential (read/)write behavior of backups.

RAID6 is now used since mostly very large (slower) SATA drives are used for the sequential behavior. RAID6 has a write penalty of 6 (!!) for each RANDOM I/O, but when sequential I/O is done this penalty is 9/7 in a 9 drive RAID6 Raid Group. The reason why RAID6 is used is because rebuilds in case of a disk failure can take very long and you wouldn't want your data to be at risk when a 2nd drive fails when the first rebuild is still running.

It's a classic case of "it depends" on your needs. Best practice nowadays is RAID6 for large SATA Raid Groups, but if you think RAID5 is sufficient enough, go ahead and use them. RAID6 is more safe than RAID5, but it's also slower.

The explanation that RAID6 is chosen because of the random I/O behavior of backups is not accurate at all, since backups tend to be sequential (I'm trynig to say this in a nice way ). The reason behind RAID6 is only data security, not performance !

0 Kudos
jps00
3 Zinc

Re: Raid 3 for backup

The recommendation to use of RAID 3 for backup performance was changed with FLARE 26.0.  In that release, the RAID 5 algorithms were optimized such that RAID 3 no longer provides a performance advantage over RAID 5 in the greatest number of workloads.  The chief advantage of the optimized RAID 5 over RAID 3 is that RAID group I/O cannot 'bottleneck' over the single parity drive.

Having written that, RRR's post discussing availability is worth consideration.  The modern backup RAID group configuration has become the 10-drive RAID 6 (8+2) which leverages the CX4-series write cache to provide both a high performance and high availability backup solution.

The EMC CLARiiON Best Practices for Performance and Availability, FLARE Revision 29.0 whitepaper describes the above in greater detail.  This document is available on Powerlink.

driskollt1
3 Argentum

Re: Raid 3 for backup

RAID6 is usually used for "Dumpster" type of boxes where they'll stripe like 45 drives into one giant RAID6 dumpster.

The largest RAID group size you can have with the CLARiiON is 16 drives.  I've typically found it more beneficial to create a RAID5 MetaLUN than to make large RAID6 RAID Group.

Your rebuild times will be faster, with less write penalty with RAID5.

RAID3 was something I've always heard about, but never really found a good enough reason to implement it.  I did some testing with it, but never really saw much improvement over RAID5.  The CLARiiON is smart enough with its caching algorithms that RAID5 would probably suit you just fine for large sequential writes.

0 Kudos
RRR
6 Indium

Re: Raid 3 for backup

> RAID6 is usually used for "Dumpster" type of boxes where they'll stripe like 45 drives into one giant RAID6 dumpster.

I wouldn't want a 45 drive RG using RAID6. A rebuild of a single failed drive would have to read data from all 44 other drives. Suppose you have 2TB drives, you'd need to read 44 x 2 = 88TB of data to be able to rebuild a single drive. If this RG is heavily used, rebuilds could take forever....

For example: I'm managing an array with 16+2 RAID6 RGs and in a heavily used RG a single drive failed (twice in 2 months). The rebuilds of each took like 4 weeks to complete. 2 weeks after the first rebuild completed, the next disk failed..... 45 Drives in a single RG ? No way !! That's for "my first storage" perhaps, but not on real arrays you're actually using in production, right ? 45 Drives in a single RG is also very bad for performance since you'd never have real full stripe writes, so all writes are considered random with a write penalty of 6 per IO.

Dumpster.... sounds cheap.

0 Kudos
driskollt1
3 Argentum

Re: Raid 3 for backup

I wasn't recommending anything, I was just stating what RAID6 is

typically used for. Cheaper boxes like NEXSAN will just hand you a

giant RAID6 stripe. For some people that's fine if all they want is a

low performance/not-many-features dumpster to keep stuff.

I typically wouldn't create a giant RAID6 stripe. I don't really like

RAID6. I think it's pretty evil, but that's just my opinion.

I mentioned rebuild times a few lines down in my original post over

RAID6 vs RAID 50. I typically use RAID50 over RAID6.

0 Kudos
jps00
3 Zinc

Re: Raid 3 for backup

> For example: I'm managing an array with 16+2 RAID6 RGs

Hmmm... a 16+2?

When considering RAID group provisioning you need to consider four separate axis: capacity, performance, availability, and logistics.

For backup with large capacity (1 TB or larger) SATA drives as the underlying storage device, either the 10-drive 8+2, or the 14-drive 12+2 have particular advantages on the CX4-series CLARiiONs.  For backup, setting the write cache to a 16KB page with these configurations abets their performance.

Using MetaLUNs are fine if you have really large capacity requirements, or if you really need to add IOPs.  However a RAID group with large capacity SATA drives can be created with a lot of capacity.  Fully cached, the performance of a single RAID group is typically enough for most backups.  RAID 6 offers the highest availability on the CLARiiON.  And the single RAID group is simple to setup and maintain. (MetaLUNs require greater expertise to setup and greater attention to detail to maintain than FLARE LUNs.)

There is a detailed discussion of the merits of the single RAID group versus other storage objects found in the EMC CLARiiON MetaLUNs: A Detailed Review whitepaper in its 'MetaLUN Alternatives' section.  In addition, EMC CLARiiON Best Practices for Performance and Availability, FLARE Firmware Release 29.0discusses the merits of the different RAID-levels in its 'RAID groups' section.  These papers are available on Powerlink.

My recommendation is to use the simplest solution that meets your capacity, performance, and availality requirements.  In addition solutions that have the lowest logistical burden should receive special consideration.  A FLARE LUN that encompasses an entire RAID group is usually the first option explored.

RRR, if you could show me that trick with a CLARiiON to get a 16+2, i'd be obliged.   

0 Kudos
Loudenj
2 Iron

Re: Raid 3 for backup

Hi All,

Thanks All for the feedback, I might suggest to EMC to update their documentation from raid 3 to raid 6. All of the EMc Clarrion Backup to disk guides for <insert product>, and the Best pratices for flare 26 - 29 still suggest that Raid 3 is way to go.

As for the I\O pattens, yes RRR, very nice way of putting it. It was their statment about those pattern that made me worry a little about the whole solution.

Again all thanks for the feed back

0 Kudos
dynamox
7 Thorium

Re: Raid 3 for backup

RRR, if you could show me that trick with a CLARiiON to get a 16+2, i'd be obliged.    

John,

RRR forgot to mention that he is Mr. Hitachi these days, anything is possible on an AMS array

0 Kudos
RRR
6 Indium

Re: Raid 3 for backup

2 USP-Vs as well as 6 AMSs

0 Kudos