Highlighted
vTao
1 Nickel

FA ports performance issue on DMX array

We have DMX3 array, currently we have 128 devices hanging off 4 FA ports, they are running 2Gbps.

Some of applications are experiencing performance issue. What are list of options we can have by improve perform on the FA ports on the arrays?

1. Migrate some devices to other less busy FA ports, which require a flat bits change, this is disruptive for application

2. Buy a new pair of 4Gbps FA ports and install onto existing array and migrate devices to the new 4Gbps FA ports, we had quote from EMC, this option is very costly.

I'd like to hear what recommendations others might have.

Thanks

Tw

0 Kudos
4 Replies
Quincy561
3 Zinc

Re: FA ports performance issue on DMX array

If your IO size is small (16k or smaller) then the CPU is likely the bottleneck and going from 2gb to 4gb is not likely to change performance much at all.

However if you are pushing large IOs and close to 200MB/sec for each port, going to 2gb may help some.

0 Kudos
s_y67
1 Nickel

Re: FA ports performance issue on DMX array

I am not concerning IO size as such. I am more concern about FA utilisation.

I am going 4Gb not 2Gb.

How easy it is to find out IO size of each application.

0 Kudos
Quincy561
3 Zinc

Re: FA ports performance issue on DMX array

I don't know what you mean by FA Utilization.

This could mean the CPU, or it could mean the pipe or plumbing.

You can look at host based stats, or FA based stats for average IO size.

If the IO sizes are large, going to a bigger pipe (4gb vs. 2gb) may help

If the IO sizes are small, going to 4gb won't make the CPU any faster

0 Kudos

Re: FA ports performance issue on DMX array

- Adding greater number of spindles on the backend might be a helpful option.

- This will help in distributing the workload across greater number of spindles, thus also ensuring a better destage and hence preventing queuing of I/O requests, which are bound to hamper performance.

- If you use thin provisioning, wide striping can be of real good help towards performance improvement.

Symm experts,

               Please validate the applicability of the above options for the given context...    

Thanks,

SreeHari

0 Kudos