RAID 10 Read Performance

1. Disk type: 600GB / SATA 15k

2. VNX type: 5300

Regarding to the bellow table, I should get excellent performance on R/W on RAID 10.

The random write speed is 131 MBs/sec which is almost 10 time faster that read.

They way I've created the LUN is: I created a RAID10 pool using all 12 disks and the I created a 3 TB LUN.

I know that the recommendation of creating RAID 10 pool is using 8 disks but the way I created is putting all disks into the pool. Could is be the root of issue?

I've enabled the performance log to send to EMC. I don't know how to examine the performance on the VNX.

Thanks,

Mohammad

Hello Mohammad,

There's a lot more data I'd like to look at in order to determine the cause of the speed difference between reads and writes, such as:

- In the random writes test, did you use the same IO size?

- Is FAST Cache enabled?

- Was the pool only recently created?

This information, and more useful points, would be available from the SP Collects and NAR files you'll gather (or already have) to contact support, which is good.

However for your knowledge; yes, its better to have RAID 1/0 RGs on a 4+4 configuration, and with high (>30%) random writes expected.

Useful document:

https://www.emc.com/collateral/white-papers/h12682-vnx-best-practices-wp.pdf

At the end of page 16:

"AVOID a drive count that results in a small remainder when divided by the preferred drive count"

Adham

Hi Adham,

1. I did the all the tests (read & write) with the same IO size

2. Looks we don't have FAST-Cache installed

3. The pool created couple of days ago.

What's the best way to create a RAID10 pool using 12 disks? I saw an article saying to create the pool with 6 disks and then add 6 more disks.

Mohammad

When you create a R10 pool, the default is 4+4. If you select 12 disks, it will create within the Pool two private raid groups:

1. 4+4 (the default)

2. 2+2 (the remaining four disks)

Ideally you would want to add disks in multiples of 8 disks (4+4). With 12 disks you get one private raid group using 4+4 and the remaining are in a 2+2. While each disk has the same performance characteristics, the individual raid groups will have a different performance level. As a LUNs is composed of slices taken from each raid group, you could see slower performance when the slices are on the 2+2 compared to the slices on the 4+4.

Each 15K SAS disk can handle between 12MB/s to 30MB/s depending on the randomness of the IO, the IO size, and the mix of reads and writes. For a 100% Read, 100% Random, 64KB IO Size, the bandwidth that each disk can handle will be closer to 15MB/s. For very random data, you'll get no benefit from pre-fetching on the array. Try changing your test to 0% Random (100% Sequential) and see what you get. Also, try 32KB IO Size.

If you have 12 disks at 15MB/s each, that means the total bandwidth should be in the range of 12 * 15 = 180MB/s. This is a raw number. It can be affected by other activity on the array, the configuration of the paths to the array, the HBA settings on the host, the failover software on the host, the load balancing on the host, and the two private raid groups within the pool.

If you only getting 13MB/s from the LUN, then there is something else occurring that is causing this. I see that you've already opened a case with EMC, they should be able to determine what is going on. It's possible that the LUN was still initializing when you ran the tests as it can take a day or more to finish.

glen