Why is more faster network between two different Hosts?

[jpesquerdo]

Hello Community!

I need help in this mystery.

I have three different VMS, all of them with e1000 (also I tried all VMs with vmxnet3) and the last vmware tools.

Two of this VMs, are in the same Physical Host (ESXi 5.5.0 2068190). Let's name for example: VM-1A  & VM-2A

Another VM, is in other Physical Host (ESXi 5.5.0 2068190). Let's name VM-1B

Both ESXi Hosts are connected with two 10Gbit NICs with load Balancing:"Route based on IP hash",  and detection:"Link status only".

Traffic Shaping is disable in all Ports.

The environment is very clean, and we don't have more VMs that can disturb the test (or inappreciable).

VM-1A: 10.12.50.201      - Physical Host ESX-A

VM-2A: 10.12.50.202      - Physical Host ESX-A

VM-1B: 10.12.50.200      - Physical Host ESX-B

Why I am seeing better network speed benchmark, when two VMs are sending something from two different Physical Host than when the both VMs are in the same Physical Host?????

It is not a "iperf" issue, I used other test with scp and I got the same speed.

Thank you very much!!!!

PS: The Physical hosts are: Dell PowerEdge R620 - 16 Xeon Cores - 384GB RAM

[SwissITGuy · Accepted solution]

Have you tried to move your VM on different hosts to the same host?

From your names:

- VM-1A on host A

- VM-1B on Hots B

ToDo

Now move VM-1B to Host-A and redo the test (both VMs on the same host). What does now happen with the speed?

Check

- How is the CPU usage during the test on Host-A, if you have 2 VMs on this host?

- What is your P2V CPU ratio? (overprovisioning?)

What it could be

- Host has not enough CPU power to serve both VMs (or overprovisioning P2V related problem). But makes for me no sense if I see your server config with 16cores.

- It is not a Host but a VM related problem (maybe it depends on VM-A2). Makes the most sense for me (your test will check this)

Please share the results

[jpesquerdo]

Thanks SwissITGuy,

When I have moved the VM to another ESX, the bad performance still.

The issue was in the RH OVF template, not in the ESX.

Thank you very much for your help.