RAID 10 vs. RAID 5...

code frog 0

So in terms of speed is RAID 10 about 5 times faster than RAID 5 is it only a little faster than RAID 5? I'm writing an article for here on an Intel NAS and I'm trying to figure out the advantages of performance of RAID 10 over RAID 5. If the consensus is that RAID 10 is faster than RAID 5 by a lot then I'll rebuild the array at RAID 10 and benchmark it against RAID 5. Metrics: 20 gig file being transferred over gigabit LAN. (Source is gigabit, switch is gigabit and destination NAS is gigabit). Presently the SATA 7200 RPM drives are set as RAID 5 with 1 spare. Transfer speed is not at all what I was expecting. (Averaging 6% of my gigabit NICs throughput according to task manager.) Processor utilization on the NAS is 90%. Is most of that due to the RAID 5 and would that improve with a RAID 10? If I convert it to RAID 10 (losing my spare) will write times be faster? Like I said, I'm writing an article for here that I think will prove to be pretty interesting and valuable so if I could some type of consensus on this it will spurn me on my merry way. It's a clients NAS that I get to play with until the servers arrive about 10 days from now. Then they go into role as NAS backup and I'd like them to be as fast as possible I'd also like my article to reflect the pros and cons of RAID 5 performance over RAID 10 or vice versa...

What I am up to: ReadyToGiveUp(Not!)[^] What friends are up to:SQLServerCentral[^]

John M Drescher

code-frog wrote:

So in terms of speed is RAID 10 about 5 times faster than RAID 5 is it only a little faster than RAID 5?

It depends on a lot of factors. RAID5 is faster at reading for most setups (since it is basically raid 0 with n-1 disks) while the writing part depends very much on the setup.

code-frog wrote:

rocessor utilization on the NAS is 90%. Is most of that due to the RAID 5 and would that improve with a RAID 10?

If the processor utilization is a result of RAID5 is then it is a very poorly designed NAS. On linux (which is what most nas devices use) using software raid 5 with a 2GHz Opteron the processor usage never goes above 10 percent for the raid5 or raid6 processing. I know that because I have 10 TB of raid(5 and 6) at work on several servers using dozens of 250 and 330 GB SATA drives on my gigabit network.

John

Steve Mayfield

This[^] article might give you some useful data on RAID variations... Steve

Millennium Knight

Use at least 6 spindles, more is even better. Use a battery backed caching controller that does HARDWARE supported RAID 1+0. Build mirrored sets of two (Raid 1) and stripe across all of the mirrored sets (Raid 0). On a SQL 2005 server I noticed a 10 fold performance increase, all other things being equal. A large, mult-database, multi-table, complex SQL query that took 10 seconds on a raid 5 took less than a second in the raid 10 configuration.

John M Drescher

Millennium Knight wrote:

Use a battery backed caching controller that does HARDWARE supported RAID 1+0.

On the one hardware controller I have in the work (3ware 9500) I used a 800VA UPS instead as it was only $100 while the BBU was like $150.

Millennium Knight wrote:

Use at least 6 spindles, more is even better.

I typically use 6 to 10 drives in my raid 5 and 6 arrays.

Millennium Knight wrote:

A large, mult-database, multi-table, complex SQL query that took 10 seconds on a raid 5 took less than a second in the raid 10 configuration.

Probably an effect of writing. Writes of smaller than the stripe size are bad on raid5 or 6 because the array has to read stripe entire stripe before writing. Pick a smaller stripe size to minimize that.

John

code frog 0

So I changed to RAID 10 the performance is double. Can you write to the RAID while it's rebuilding as 10? It sure looks as if you can...

What I am up to: ReadyToGiveUp(Not!)[^] What friends are up to:SQLServerCentral[^]

Lost User

Software or hardware RAID controller? It makes a big difference.

Visit http://www.readytogiveup.com/[^] and do something special today.

Lost User

Use volume nanagement and if you have 8 drives then 2 arrays of 4 drives keeps the rebuild time down.

Visit http://www.readytogiveup.com/[^] and do something special today.

Andy Brummer

The benchmarks that Steve posted seem to backup what I would have predicted, a RAID 10/0+1 performs about as well as a RAID 0 with half as many drives, since the size of the RAID 0 portion of the array is what increases performance. Basically the greater the number of spindles that you can throw at individual IOs the better. It also backs up the higher cost of dealing with parity in RAID 5. One thing that might be interesting to test is how a RAID 10 array performs vs. a RAID 5 array with a failed drive. I've heard that the extra redundancy of the RAID 10 really shines since it just reads off the good disk while the RAID 5 array has to reconstruct the data from parity information. I'm sure it depends on the actual RAID implementation though.

This blanket smells like ham

El Corazon

Trollslayer wrote:

hardware RAID controller

cached RAID controller, cached and parallel access RAID controller, interrupt driven? Intelligent optimized reading/writing (command queuing)? pre-cached reads? multi-user? single-user? SCSI? SAS? SATA I? SATA II? The list of questions can go on and on and on. If it didn't then all RAID controllers would perform exactly the same and there would be no room for vendor variation and improvements. :-D

_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)