Very slow network perfromances between VMs (vpshere 5 cluster)

[aleber76]

Hi everybody

I'll try to describe my setup and issues:

One cluster with two hosts, ESX 5. The hosts are two HP Proliant DL 385 G7, two AMD opteron 6134 processor each, 16 cores per host.

Every host has 70GB of RAM. There are 4 gigabit nic connected per host.

The host are attached to a HP MSA 2000 G3 SAS, the LUN assigned to vms datastore is a RAID 10 made by 8 SAS disks (300GB 10k).

Our biggest problem is that VMs in this cluster are experiencing verly slow network transfer rate between each other.

On the other way, If I try to perform a test between a physical server and a VM in this cluster (every newtork card is attached to a gigabit HP switch), i get reasonable results.

If I perform a test between VMs, i always get a value range 50 - 100 Mbps. (on many other "similar" installations I never get values below 200mbps...)

All the VMs are Windows 2003 R2 32bit, vm hardware version 8, VMXNET3 installed, all offload options disabled.

This is driving me crazy...even with my lab host (an old HP server with a raid1 and 5 vms), I get better value in network transfer between vms!!

Please, any idea or suggestions are wellcome!

Thanks in advance

Alex

[aleber76]

Sorry, I forgot to mention that I already upgrade SAN firmware to the latest (2.30, that supports VAAI providers), and installed VAAI drivers on both hosts.

[TheITHollow]

how are the networks between the 2 vms related?

are they both on the same vlan? or do they need a router to pass packets between the two?

If they are in the same portgroup, and the vms are on the same host, the traffic should never have to leave the host and might shed some more light on this.

if the traffic is traveling between hosts, or subnets maybe there is a bottleneck elsewhere?

you could try a product call Xangati which is free for 14 days I think.  Setup your vDS to send netflow to the appliance and it might give you more in depth data.

Good luck!