Hey All,

AFAIK this is my first post on the community forums. I generally spend lots of time on the VMware IRC channel. As I'm not aware of a EMC based one. Anyway I'll get my infrastructure information out of the way so we have a baseline.

I'm running a VNXe3150 with the latest firmware. I have been pretty displeased with this box since we got it, but I won't get to the many points that brought me to that conclusion I'll just get what I've observed.

After much research I decided to move from NFS to iSCSI, mainly for MPIO so I could have increased svMotions when I need to make infrastructure changes. I did lots of networking redesign to free a bunch of ports on my Cisco 3750 (stacked) core layer 3 switches. I created 2 new vlans (22, and 24). I used the VNXe built in nics Eth 2 and 3 (on each SP respectively)  giving them each an IP as required (SPA iSCSI A 22.1 and 24.1) (SPB iSCSI B 22.2 and 24.4).

I then created 2 new vSwitches on each of my ESXi 5.5 hosts, each with their own VMKport bound to one NIC each. (Host 1 22.200, 24.200) (Host 2 22.201, 24.201). I then verified each hosts software iSCSI initiators where set.

Now in Unisphere, I have my hosts added to it, so Unisphere can configure my new datastore on my host when I create a VMware based datastore. There are a couple thing however that I noticed. It does add the static targets to the initiators and re scans and mounts the datastores; however.

1) I have to convert them to VMFS 5 from on of the hosts Storage section settings

2) I have to set the load balance algo for the managed paths from Fixed to Round Robin.

Once I did this all, I was able to verify that MPIO was working and the load was being distributed equally. I wasn't too impressed with the increase performance since instead of hitting a 1 gbps limit as with NFS, I was only able to hit 140-150 MB/s, this is roughly only 55% of both gbps NICs. Even with this I expected a increase in VM performance for IOPs. To my dismay it was not what I had expected.

See Test 1, VM on Hosts local Datastore, consisting of 6 - 300 Gig 10K SAS drives in - Pastebin.com

I kept testing and trying things, then people on IRC told me about the Round Robin setting to change from iops=1000 to iops=1, which basically tells the host to load based on each iop instead of every 1000 IOP. After I made this change my mind was Blown. However my VM performance on the iSCSI datastore remained the same.

What am I missing? (I already read the posts about NFS cache, when I check the settings under advances settings, cache is DISABLED)