Solved!
Go to SolutionRAID5(4+1) vs RAID5(8+1)
We are having a discussion if we should go with raid5(4+1) or raid5(8+1).
Other than space saving advantages with (8+1) what else do we gain and I know there is going to be a performance impact if we go with (8+1) but what else are we going to loose.
in a 41 raid group you'll get 125% of the IOps per GB of a single drive in the RG where in a 81 you'll get 112.5%.
In performance calculation you'll always count ALL drives, which is fair snice the write penalty dealt with the overhead.
Traditionally I'd have gone for creating a META LUN over 2 RGs which both are 41 rather than 1 LUN in a RG which is 81. Twice a 41 has the performance of 10 drives, where a single 81 has only the performance of 9 drives.
We are thinking about RAID5(8+1) because it will save us a lot of disk when compared to RAID5(4+1).
Not sure if it is the right move but want to know if there is a way in Unisphere to create a RAID5(8+1)? I think it defaults to (4+1).
Thanks.
Storagesavvy
474 Posts
3758
0
Posted March 4th, 2011 08:00
A RAID5 Group of 10 Disks will create a 9+1.
A RAID5 Thin Pool of 10 Disks will create two 4+1’s.
Oh yes that was just me being dumb. We can choose as many disks as you want and the system will RAID it depending on that.
So my understanding is that when you create a RAID group with 9 disks, we are create a LUN with size 500GB, are we spanning the parity accross all the disks? Is that same as creating RAID5(8+1)?
If we created a pool with 10 disks how does it lay out in the backend? Is it going to be RAID5(4+1) or (8+1)?
Appreciate it.
Allen Ward
6 Operator
•
2124 Posts
3758
0
Posted March 4th, 2011 11:00
So my understanding is that when you create a RAID group with 9 disks, we are create a LUN with size 500GB, are we spanning the parity accross all the disks? Is that same as creating RAID5(8+1)?
Just to address this specific point, when you create a RAID Group with 9 drives it does not yet have any RAID type associated with it. When you create that 500GB LUN you define what RAID type you want it to be. You could make it RAID 5 which would distribute the parity in an 8+1 configuration. You could also choose RAID 6 which would distribute the parity in a 7+2 configuration. There are other options as well. The point is, nothing is set for the configuration until you bind your first LUN on the RG.We have 3 SATA II SSD disks. There are different size than our fiber channel disks.
I am thinking it will not be a good idea to mix disks of different sizes in the same pool. Am I correct or is there a reason to use SSD disks in combination of fiber channel disks even if they are different in size.
jps00
2 Intern
•
392 Posts
3755
0
Posted March 1st, 2011 10:00
I recommend reviewing the EMC CLARiiON Best Practices for Performance and Availability, FLARE Rev. 30.0's Rebuild and Stripe Size sections. This document is available on Powerlink. However, consider performance and availability of small versus large RAID groups.
From a performance standpoint, the stripe size of the larger RAID group will abet large I/Os. Backup likes the 8+1 and 8+2 stripe. You also have a greater number of IOPs in that larger RAID group.
From an availabity standpoint, an 8+1 can take a long time to rebuild, especially if its near capacity. You also need to consider the availability 'failure domain' aspect. An 8+1 has a higher likelihood of a drive failure than an 4+1. Also there can be a lot more in-use capacity with a 8+1 than a 4+1; consider the effect of a double failure. See the EMC CLARiiON Storage System Fundamentals document for a discussion of Mean Time Between Failure (MTBF) and calculating the likelihood of a RAID group drive failure based on a 100K hour MTBF. This document is available on Powerlink.