Skip to main content

What's the performance impact if we change RF from 2 to 3?

Dears:

We have a 3 node pure NVME SSD clusters with RF 2. Now we plan to expand more 5 nodes and change the RF from 2 to 3.

 

We have a little worried about the performance on :

(1). During conversion phase from 2 copies to 3 copies.

(2). For BAU usage, if RF3 performance will slower than RF2 because every write need do 3 times compared of 2 times.

 

I got the question 1 answer from Nutanix Resiliency – Part 2 – Converting from RF2 to RF3 | CloudXC (joshodgers.com) and want to see any more experience.

 

For question 2, any experiences could be shared?

 

Thank you very much!

 

P.S. We have enabled compression feature.

Hello @LinKan 

From Read prospective, There is no performance impact

From Write prospective, you are using NVMe SSD, you will have 1 tier in the extend store, 1st copy will be written locally, then another 2 copies will be sent to 2 another nodes. The main factor here in the performance is the Network configuration, I recommend 25GB ports.

The most important question is, what is your cluster I/O latency now at RF2? 

@Moustafa Hindawi : We use 40GB ports for network traffic within Nutanix clusters.

From overall cluster level response time, we can see less 1 ms except some high response time which is caused by SQL backup and restore period.

 

I have another question for “How Nutanix deifne the write complete”: will Nutanix treat host write operation complete until two copies are written or just local copy is written. I think it is key for performance impact.

 

 

Nutanix treat host write operation complete until two copies are written.

Hello @LinKan 

From Read prospective, There is no performance impact

From Write prospective, you are using NVMe SSD, you will have 1 tier in the extend store, 1st copy will be written locally, then another 2 copies will be sent to 2 another nodes. The main factor here in the performance is the Network configuration, I recommend 25GB ports.

The most important question is, what is your cluster I/O latency now at RF2? 

@Moustafa Hindawi : We use 40GB ports for network traffic within Nutanix clusters.

From overall cluster level response time, we can see less 1 ms except some high response time which is caused by SQL backup and restore period.

 

I have another question for “How Nutanix deifne the write complete”: will Nutanix treat host write operation complete until two copies are written or just local copy is written. I think it is key for performance impact.

 

 

Hello @LinKan 

From Read prospective, There is no performance impact

From Write prospective, you are using NVMe SSD, you will have 1 tier in the extend store, 1st copy will be written locally, then another 2 copies will be sent to 2 another nodes. The main factor here in the performance is the Network configuration, I recommend 25GB ports.

The most important question is, what is your cluster I/O latency now at RF2? 

Terms & Conditions