Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

I understand about the different types of RAID configurations:

  1. RAID 1 -- Mirror [Redundant]
  2. RAID 0 -- Stripe [Performant]
  3. RAID 5 -- Hot Spare [Redundant]

From what I understand, RAID 1 gives you redundancy with no performance hit, RAID 0 gives you performance at the cost of redundancy, and RAID 5 is good in the senes that you gain extra storage space and still have redundancy, at the expence of performance.

I also understand a bit about Nested RAID configurations, specifically RAID 10 (1+0). This provides the benifit of both RAID 1 + RAID 0. Giving you the best of both worlds.

My question is this: (And a SAN is out of the question here) Would a RAID 50 or a RAID 10 be better suited for a virtualization environment where most VMs are very under used in terms of disk IO and CPU use. I'm trying to maximize the number of VMs I can put on a host machine, and my limiting factor is drive space.

So, I'm trying to put as many VMs on a Virtual Host as possible, most of the IO on any of the VMs will be OS related and on occasion user requests will trigger a VM to do some IO, but this is rare and relativly evenly distributed across VMs.

My thought is that the RAID 50 could prove better because A) better use (2/3 total size of all disks in the array instead of 1/2) and B) potentially better better random read/write performance since there are more drives in play for storage, so read/write seeks are more likely to be on different physical drives.

update

Virtual Hosts - Windows Server 2008 R2

Virtual Machine - Windows 2000 Server - Windows Server 2008 R2

update 2

Single hardware RAID controller. Which I assume would do parity calculations?

Two RAID 5 arrays < 10 total drives

Raw performance is not an issue in most situations, the main limiting factor is physical drive space. Most of the IO is just OS chatter. Due to limitations in our database software we need a 1-server::1-system, even if its only 5 users. So most of our database servers sit idle. The overhead is just running that many OSes on the drive array.

update 3

I am currently using a RAID 10, and it is working just fine; but I'm finding that more drive space would be useful, (thus the inquiry about RAID 50 for better use of physical disks). As suggested below, I'm hearing that RAID 6 could offer similar performance, redundancy, and storage capacity as RAID 50 -- is that true or am I not understanding the responses?

share|improve this question
    
which OS are you looking for this setup!! –  Rajat Aug 3 '10 at 16:30
    
Updated question body and re-tagged, thought I'm not sure how it's relevent. –  Nate Aug 3 '10 at 16:39
    
More relevant: if you do RAID 50, how many RAID 5 arrays and what will be doing the parity calculations? Do you have one RAID card for each underlying RAID5, one RAID card for all the underlying RAID 5 arrays, planning to do it all with software RAID? –  freiheit Aug 3 '10 at 16:50
    
@freiheit Updated question, I'll have one RAID controller (hardware), < 10 drives total, I presume the RAID controller will do the pairty calculations? –  Nate Aug 3 '10 at 17:00
add comment

2 Answers 2

up vote 1 down vote accepted

Honestly, the answer to your question depends, in part, on what the rest of your installation looks like. You're asking questions about redundancy levels on local RAID controllers when you're talking about storage that itself has a single point of failure, the host machine that it's connected to.

Unless you go to some sort of shared array (which might be much simpler and lower-end than a SAN) you'll find that failure of your physical compute host will take down all your workloads. This will always be the case unless you move to a shared array that doesn't use your local RAID controller. This fact may lead you to a strategy where you back up your VMs pretty often, more so than you might if you clustered two or more hosts with a shared back end.

And if you do those backups so often that you can survive a host failure, you might conclude that you just want to use that local RAID controller in a striped (RAID 0) mode for pure performance.

Honestly, though, in your situation I'd probably be looking for a single RAID6 array, if your controller supports it. As "freiheit" said, though, it depends very much on the type of storage traffic that your VMs are going to generate. A lot of random writes from several VMs simultaneously might overwhelm your RAID array. If so, you'd want to optimize for performance. If not, you might want to optimize for reliability, depending less on backups.

share|improve this answer
    
Could you provide a bit more information on what type of shared array, that isn't a SAN, that would provide acceptable performance for VMs? I realize that the RAID controller is a single point of failure, I'm looking at RAID for performance and data integrity reasons first and foremost. –  Nate Aug 5 '10 at 15:46
    
Sure. Shared SAS arrays tend to be about half the cost of "SANs" though some people will call them SANs. The distinction is that the shared SAS array, even if it comes with a switch, it extends no further than one or two racks, typically has somewhere between 2 and 8 hosts attached to it and it provides RAID in the array, not the HBA. A Fibre Channel or iSCSI SAN, in contrast, would provide thin provisioning, many more hosts, LUN snapshots, LUN migration, etc. Cost of an array has everything to do with the features it supports. If all you want is RAID, you might not need the high end. –  Jake Oshins Aug 5 '10 at 18:54
add comment

How many drives are you talking about? How much storage do you need? Are you talking about 2 RAID5s in a RAID0, or are you talking 20 RAID5s in a RAID0? What's the hardware? Fundamentally, the performance difference between RAID50 and RAID10 depends on the hardware.

RAID 1 gives you no performance change on write, but may double your read performance (especially with multiple simultaneous read requests; they can be spread across both drives)

RAID 5 isn't "hot spare", it's more like "striped parity". You can lose any single drive in a RAID5 and it continues to operate, but in a "degraded" fashion. Write performance to a RAID 5 can be limited by how fast you can calculate parity; a given RAID card will have a maximum speed that it can calculate parity and therefore a maximum write speed, regardless of how many drives you have.

A RAID 50 (RAID 0 of RAID 5 arrays) could work well. It's similar in performance to a single large RAID 5 or a JBOD of RAID 5. However, a RAID10 has no parity calculation limit. Unless you've got some beefy hardware, it's likely that parity calculation on the multiple underlying RAID5 drives will be a limiting factor on writes. It really depends on what's doing the RAID5 parity calculations.

Whether you do RAID 50 or RAID 10, with that many drives you want to be sure to keep a hot spare (an unused drive ready to be moved into any of the underlying arrays in case of drive failure). With that many drives, it's inevitable that one will fail and a hot spare can buy you time. You might also consider RAID6 (RAID5 double parity) instead, since it's more fault tolerant. Performance, space usage and fault tolerance of a single RAID6 should be very similar to a RAID 0 made up of two RAID 5 arrays.

share|improve this answer
    
In my situation it would likely be 2 RAID5s + RAID0. But you are saying that a RAID6 might be similar in performance to a RAID 50; but neither quite as fast as RAID 10? –  Nate Aug 3 '10 at 16:52
    
I'm saying that one bottleneck with a RAID 50 (any RAID5 or RAID6, really) is parity calculations when doing writes. Whether that's a more important bottleneck than the actual write performance of the drives depends on how fast the drives are and how fast the RAID card can do parity calculations. In other words, it's not really a "general" question, it's very specific to the very specific hardware. –  freiheit Aug 3 '10 at 17:10
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.