Raid

[LukeC]

I'm purchasing a new server to be a hyper v virtual host with local storage. I'm undecided what raid set up to go for either Raid 5 or Raid 10 but a supplier has just mentioned about Raid 50. It will host around 4 virtual servers on average with an extra one or two in the event of a disaster, some of the roles will be exchange, file services, WDS/WSUS, Website and another SQL application possibly. It would be much appreciated if you could cast your opinions and advice.

[sted]

looks like a way of selling more drives 6 drive minimum 3 lots of 2disks in raid 0 then raid 5'd together i also suspect it will be a fair load on the raid card. Not sure what it achieves that a 3 disk raid 5 with a hotswap dosent do

[VeryPC_Ed]

I'm purchasing a new server to be a hyper v virtual host with local storage. I'm undecided what raid set up to go for either Raid 5 or Raid 10 but a supplier has just mentioned about Raid 50. It will host around 4 virtual servers on average with an extra one or two in the event of a disaster, some of the roles will be exchange, file services, WDS/WSUS, Website and another SQL application possibly. It would be much appreciated if you could cast your opinions and advice.

 

Hi Luke, I would suggest looking at Raid 10 as this will provide the best mix of performance and redundancy, especially if you are running i/o intensive applications such as SQL and Exchange.

 

I would also suggest tiering your storage as you could look at slower larger capacity drives for your non-critical files such as WDS/WSUS and possibly your general file storage (dependant on users).

 

I'd be happy to provide some different options based on your requirements, i'll send you a PM with my details and what info I would need to size up a spec suitable for your needs.

 

Thanks,

 

Ed

[AHeighton]

Personally, I'd look at RAID6 if I were you.

RAID5 was fine when the chance of a drive dying was pretty low and the time it took to rebuild an array was short. Now drives are pushed so close to the limit of what's possible, the chance of seeing a single drive fail when you have 6 of them is more like 50% for a given year.

 

RAID6 has two recovery volumes rather than one, which protects you against the scenario where one drive fails, and then during rebuild, another fails too. With RAID5 you've lost everything that hasn't already been rebuilt, with RAID6 you can continue the rebuild.

 

Of course, that means having 4 drives puts you at data/data/recover1/recover2 = 2/4 = 50% drive space available to use

Or with our 8 drives in RAID6, that's data/data/data/data/data/data/r1/r2 = 6/8 = 75% drive space available to use.

 

A RAID50 configuration doesn't give you as much protection as you might expect - a single drive can fail in either half of the pair, but if two drives fail in the same half, you're still just as screwed. And you've only had 50% of the drive space available to you in the first place.

 

So RAID50 bad, RAID6 good, in so much as available space and fault tollerance goes anyway.

 

If your card can't do RAID6, you're being sold a dud card.

 

Alex

[Snuffkins]

Personally i'd go for RAID 10. It's a trade off of capacity for performance, but also gives you reasonably good protection against disk failure, especially with a hot spare. The re-build process should be faster and shouldn't inccur such a performance hit during the re-build. RAID 5 and especially RAID 6 rebuilds can take quite some time to achieve. It's more of a problem with RAID 5 though as if you get a second disk fail during the rebuild you've lost the array. The problem with the rebuild for RAID 6 is the performance impact it has.

 

Looking at the VMs it will be running though, the performance benefits of RAID 10 may be a little wasted. If i was picking between RAID 5, 50 or 6 in that case i'd pick RAID 6. I would never really use RAID 5 anymore, although it doesn't suffer from quite the performance hit that RAID 6 does, you really are trading off protection for not much of a performance gain. RAID 6 can withstand 2 disks failing at the same time because it doubles up it's parity data. RAID 10 can also withstand 2 disk failure provided they do not occur in the same mirror.

 

While RAID 50 is as an option, i don't think i'd go down that route with the others available although it is quite a good balance of performance, cost and capacity i suppose. If you do, just like with RAID 5 the less disks in each array the better your chances of not losing the whole lot to multiple drive failure in the same array., but only because there are less drives to fail.

[tmcd35]

looks like a way of selling more drives 6 drive minimum 3 lots of 2disks in raid 0 then raid 5'd together i also suspect it will be a fair load on the raid card. Not sure what it achieves that a 3 disk raid 5 with a hotswap dosent do

 

That's RAID-05, RAID-50 is the other way round (two RAID-5 arrays Raid-0'd together). Besides I thought there was a 4 drive min on RAID 5 (could just be my ailing memory getting it wrong).

 

 

6 VM's ain't an awful lot and RAID-50 does sound a bit overkill. RAID-6 is in option for extra redundancy but finding compatible controllers might had to the cost. RAID-10 is good for redundancy and speed but is wasteful on storage space. Personally, I'd probably opt for RAID-5.

 

But before we can advise properly we do need to know two things - what services are the VM's hosting, and what size/type hard drives are you planning on using?

Edited June 13, 2013 by tmcd35

[Guest Guest]

Personally, I'd look at RAID6 if I were you.

RAID5 was fine when the chance of a drive dying was pretty low and the time it took to rebuild an array was short. Now drives are pushed so close to the limit of what's possible, the chance of seeing a single drive fail when you have 6 of them is more like 50% for a given year.

 

RAID6 has two recovery volumes rather than one, which protects you against the scenario where one drive fails, and then during rebuild, another fails too. With RAID5 you've lost everything that hasn't already been rebuilt, with RAID6 you can continue the rebuild.

 

Of course, that means having 4 drives puts you at data/data/recover1/recover2 = 2/4 = 50% drive space available to use

Or with our 8 drives in RAID6, that's data/data/data/data/data/data/r1/r2 = 6/8 = 75% drive space available to use.

 

A RAID50 configuration doesn't give you as much protection as you might expect - a single drive can fail in either half of the pair, but if two drives fail in the same half, you're still just as screwed. And you've only had 50% of the drive space available to you in the first place.

 

So RAID50 bad, RAID6 good, in so much as available space and fault tollerance goes anyway.

 

If your card can't do RAID6, you're being sold a dud card.

 

Alex

 

 

^ You sort of understand raid levels but you don't seem to know how they apply to real world usage. For example you would never use RAID 50 in a 4 drive array*, firstly because its impossible as you need a minimum of 6 disks, and secondly because its for large arrays. To say raid 50 = bad, raid 6 = good is just wrong. In a low write storage array such as shared file storage then the statement is sort of true, but in the OPs case of a high write, high IOPs situation RAID 50 is far far better than RAID 6.

 

*Reason I say 4 disk array is you quote RAID 50 as losing upto 50% of its storage capacity, which again isn't true so I'm not really sure how you are working things out here.

 

 

Anyway, to the point;

 

RAID 5 and especially 6 is not really suitable for VM hosts, especially with exchange and sql instances on, due to the slow writes.

 

As a default position I'd go raid 10 but without knowing your needs... You might find that RAID 50 is your best option due to your storage requirements vs number of drive bays.

 

How many bays do you have? Split storage as suggested above may be good. A dedicated RAID5/6 array for your file server and WDS/WSUS may be a possibility depending on how many users/are profiles stored there/etc. The RAID 10 for all OS drives and other VMs. One interesting thing to mention is that exchange can actually be better at handling the striping than a RAID card. Some people set up a load of mirrored arrays and then present them all to exchange which stripes them. As exchange is aware of the data in the stripes (where as a raid card only knows it's 1s and 0s) it can be more efficient about how it stripes the data.

Edited June 13, 2013 by Guest

[tmcd35]

Just a note on RAID-50, We run our entire school network off the back of 12x450Gb 15k SAS drives in RAID-50 + 2 drives as hot spares + 2 drives mirrored for the hosts OS. We run around 30 VM's off it plus all the homes drives, etc. Never had an issue with speed. In theory I can have up to 4 drives fail without data loss. We've had a couple of drives fail in the past and the Hot Spares auto rebuilding is brilliant!

[LukeC]

I was thinking of using 15k 600GB SAS Drives.

[AHeighton]

To say raid 50 = bad, raid 6 = good is just wrong. In a low write storage array such as shared file storage then the statement is sort of true, but in the OPs case of a high write, high IOPs situation RAID 50 is far far better than RAID 6.

I accept that RAID6 is slower in some scenarios, yes, but personally I wouldn't dare risk it. The chance of a second disk failing during the long rebuild would just scare me too much!

 

*Reason I say 4 disk array is you quote RAID 50 as losing upto 50% of its storage capacity, which again isn't true so I'm not really sure how you are working things out here.

Also, yes sorry, I was thinking RAID51, which loses 50% for the mirror and then 1-(n-1)/n for the redundancy.

 

I think it depends on your personal strategy on catastrophic failure. A striped array made out of RAID5 arrays is understandably nippier than the Belt-and-Braces RAID6, but again, if two disks fail on the same side of the stripe, you've still lost all your data. You can put the same number of disks into a RAID6 and get the ability to continue functioning if any two disks fail, whether they're both in the first half of the bays or the second.

 

Whichever OP picks, it's clear you would go 50 and I would go 6.

 

I think I know who trusts their backups most!

Edited June 13, 2013 by AHeighton

[Guest Guest]

I was thinking of using 15k 600GB SAS Drives.

 

How many?

 

How many users?

 

What's this SQL instance going to be for?

 

Do you have a separate server for profiles?

 

How many computers does wsus serve?

 

How many computers do you rebuild concurrently using wds?

[Guest Guest]

 

Whichever OP picks, it's clear you would go 50 and I would g

 

I think I know who trusts their backups most!

 

Actually I pretty much always go raid 10. The only reason I'd go raid 50, 5 or 6 is for low write storage such as wsus, wds, shared areas or backups. But then that's based on my schools usage - high number of users = massive file storage needs so no raid 5 for that. Fairly high concurrency so no RAID 6 for storage either. DBs etc are high writes/IOPs so only route there is RAID 10. OSs on raid 1 or 10. So in simple terms, RAID10.

Edited June 13, 2013 by Guest

[Michael]

As you can tell from this discussion, there is no right or wrong answer. All the suggestions/reasoning are good, but I would narrow it down to RAID 6 or RAID 10. RAID 50 is also good, but very costly. You lose a lot.

[Guest Guest]

^ RAID 10 you lose a lot not 50. But you are right, it's down to your specific needs not RAID x > RAID y.

 

In oder from most wasteful to least.

 

RAID 10 = n/2

RAID 60 = n-4

RAID 6 = n-2

RAID 50 = n-2

RAID 5 = n-1

[tmcd35]

@j17sparky, Beat me to it - lol.

[tmcd35]

I was thinking of using 15k 600GB SAS Drives.

 

The other thing to consider with RAID is more spindles = more speed (lower seek/access times). So depending on your enclosure using say 8x300Gb drives might be better than 4x600Gb drives which also opens up the RAID options into the RAID-50 or RAID-60 territory. Also if you are thinking of a large array, in terms of disk numbers, then seriously consider 1 or 2 hot spares.

[Guest Guest]

^ yep. To further that you can consider 2.5inch over 3.5inch as you generally get around twice as many disks. You generally only get 10k disk but it is negated by te double spindles. Plus it gives you plenty of room for hot spares.

 

@LukeC arnt you glad you opened this can of worms?

[apeman]

The other thing to consider with RAID is more spindles = more speed (lower seek/access times). So depending on your enclosure using say 8x300Gb drives might be better than 4x600Gb drives which also opens up the RAID options into the RAID-50 or RAID-60 territory. Also if you are thinking of a large array, in terms of disk numbers, then seriously consider 1 or 2 hot spares.

 

Not sure i agree with the more spindles statement, with certain raid levels yes more drives can/could be quicker but again it all depends on the data being written/read from them.

EMC always recommend Raid 5 4+1 for performance anything else is considered slower in there setup.

 

For the OP my personal choice would be raid 10 or 50. For raid 6 you need a really good raid card as it has to calculate the parity bit twice which requires a lot more cpu cycles.

[ConradJones]

Hi Luke, I would suggest looking at Raid 10 as this will provide the best mix of performance and redundancy, especially if you are running i/o intensive applications such as SQL and Exchange.

Ed

 

All our production stuff runs off raid 10 LUNS i have two raid 6 luns for the testing kind of stuff

[mister_z]

For a database that is read-mostly, RAID 5 or 6 is fine. My organization uses it all the time for low write/delete/modify databases. And we use RAID 6 for much of our data warehouses (tempspace and index rebuild areas excepted). RAID 5 is only slower during write ops, not read ops. So consider the usage.

 

RAID 1 and it's big brother RAID 10 we use for fast transactional systems, beginning with the log space (since that's written to with every write op), then for the temp space (large sorts, etc), then as needed for high change rate tablespaces.

 

For spindle capacities > 600GB, we go to RAID 6, since the hotspare rebuild times start to get longer, and we really don't want to lose another drive while one is rebuilding. Else it's "Hope the backup set is current," and "Tell the users it'll be down till the restore finishes."

 

Also, don't confuse throughput/bandwidth with access time. They only correspond when there are enough users active, and then only roughly. Without a demonstrated bottleneck, adding more spindles to a RAID set (thus raising the stripe size) means more users can get to data in the same time, but not necessarily that the same number of users can get to the same data faster.

 

Lastly, remember the hot spares. Our rubric is 1 HS per 30 disks in service if the hot spare can take over for any other spindle in the same VNX or VMAX (my teams manage several data centers with rather sizeable SANs in each). If you do not have hotspares, you MUST HAVE ACTIVE MONITORING watching for failures. I have seen too many organizations lose a disk in a redundant system and not know it until a second one fails. "Oops, this is going to suck."

 

As I recently explained to a colleague in another division, my preference (not always possible) is to start as small as possible and "watch and grow" the system as performance data becomes available. I advised him to keep 30% of his space in reserve and adjust its placement as he learns more.

 

Best of luck.