VMware Cloud Community
alpferd
Contributor
Contributor

Slow network when VMs on same host

Hi all,

we're running 2 hosts with ESX 4.0.0 164009 within HA+DRS cluster.

If the virtual machines are splitted over both ESX hosts, the connection is very fast (download ~1 MByte/s).

But if all virtual machines were on same ESX host, the connection is very slow (download 4-100 KByte/s)

I've tested this with 2 and 3 systems in a chain (see attachment)

Example 1:

RDP connection from internet to windows system in DMZ.

The connection is working, but you can see blocks during window change. (very slow!)

Example 2:

Network connection from Internet via transparent packet filter to

the reverse proxy where public ip is NATted to internal webserver ip.

Internet => Packet filter => reverse proxy => internal webserver

Why

NAT on reverse proxy? We have disabled reverse proxy functionality to

simplify transport and exclude reverse proxy as root cause.

Do you have any idea what could be the root cause?

IMHO network should be faster while traffic is only on local host and not leaving ESX host.

I've played with several adapter types (e1000, vmx2, vmx3), but there isn't any change.

packet filter: 3GHz, 2GB RAM, RedHat EL 5.3 x86_64 2.6.18-164.el5 with iptables working in transparent mode

windows system: 3GHz, 4GB RAM, win 2k3 32bit

reverse proxy: 3GHz, 2GB RAM, RedHat EL 5.3 x86_64 2.6.18-164.el5 with iptables

internal webserver: 3GHz, 4GB RAM, RedHat EL 5.3 x86_64 2.6.18-164.el5 with apache

Regards

alpferd

Tags (4)
0 Kudos
1 Reply
alpferd
Contributor
Contributor

Yeeeeeeeeeeaah!!!! solved!

http://rhymingpanda.com/weblog/2007/03/13/20_22_12/index.html

Set the tcp segmentation offload to off:
$ sudo ethtool -k eth0 \\ rx-checksumming: off \\ tx-checksumming: off \\ scatter-gather: off \\ tcp segmentation offload: off$ sudo ethtool -K eth0 tso off \\

I set this on all VMs and network speed is back!!!! Smiley Happy

0 Kudos