S2D performance: 2-way mirror vs 3-way mirror

How many mirrors?

Over the past few months we’ve deployed several Storage Spaces Direct solutions, and one question that seems to pop up is the expected performance difference for a 2-way mirror vs a 3-way mirror. The obvious answer is a 2-way will out-perform a 3-way, but by how much?

I won’t go into the resiliency or design decisions or when either option is preferred, this is purely from a performance based perspective..

 

The chosen system for this test was a fairly cost effective low-end system.

4-node Hyper-Converged. Each node being:
2x Xeon E5-2640 v4 (10C/20HT)
128 GB Memory
2x 400GB SSD in RAID1 for OS
4x 1.92TB SSD SATA 6Gbps
8x 6TB 7.2K SATA 6Gbps
1x Mellanox ConnectX-3 Pro

 

The tool of choice responsible for our S2D stress testing was none other than VMFleet.

We could do a myriad of tests here, but to keep things simple I used the below settings for our testing. The theory here is to simulate the average workload

Byte Size: 4K
Threads: 4
Queue Depth: 20
Read/Write: 70%/30%

First, the 2-way mirror – approx. 660K IOPS

image

Now, the 3-way mirror – approx. 520K IOPS

image

*these tests were ran over a period of 30minutes and the above indicate the average results.

So in this particular deployment, we see approx. 22% performance overhead by using a 3-way mirror. Not that much of a hit considering the resiliency benefits. But again, this is dependent on the deployment..

Don’t use this data as a constant for all environments, it was posted simply to show that a 3-way mirror doesn’t impact performance on a grand scale, just a little bit Smile

For anyone speculating the performance characteristics of a 2-way vs 3-way mirror, I hope this helps in some way

Enjoy!
Dan

Leave a Reply

Your email address will not be published. Required fields are marked *