need more info, what settings did you use in IOmeter. IO size, read/write ratio, seq / random ..etc

Also the number of threads is needed...FC or iSCSI connection...

did you mean to attach something ?

Not sure what happened.

Here is my response

This is fiber connection.

Using 4k 100% write random

2000000 sectors

64 targets per io

The results are changing.

Now I am getting 20000io/s

75mb/s

Av ioresp 3ms

Max io resp 476ms

I had to remove the rdm from the vm. Shutdown the vm and readd to restart any iometer tests. Why?

Are these acceptable results?

Not sure why. It starts out at 25000iops and then drops down to 1500 iops

There are a lot of variables here.  Can one assume from the fact that you are performing IOMeter tests that the following is true:

1. ESX Host has not other active VMs?
2. The SAN has no other I/O Load?
3. The Fiber Fabric has no other I/O Load?

Assuming all that is correct, can you look at the performance logs for the array and see any bottlenecks at the LUN. RG, or SP sides?  On the host, can you tell us if you are getting any Queuing and what the number is?

A few other questions:

1. Are you using caching or is it diabled for the test?
2. Are you using FAST Cache?
3. You say nothing is "really on" the RG?  Is there other stuff there at all? If there is other data there, is it fragmented?
4. Can you tell us the speed of the disks, the makeup of the RG, etc?

Happy to help if you can provide more data.

Thanks.

First the write cache fills up and you'll get a really good IO rate. After the write cache is full, the IOps drop down to a steady state, which in your case is 1500. In the background the write cache will do flushings to disk and on the front end your test machine is filling up the cache, so this 1500 is the actual speed you're getting from the disks, since they are the weakest link.