VMware Cloud Community
barondavid
Contributor
Contributor

Slow network speeds on ESXi

Hi we have had continuous network speed problems with our host an running out of ideas.

First VM's just about 10/10 Mbit speeds, but we saw we had some auto negotiation problems between the ESX and our switch. We have now tried with both fixed 1000 full duplex and with another NIC where auto negotioatin works, this gives better speeds, but still strange results.

When trying to test our internet speed on VMs we get 80-130Mbit down and 200-250 Mbit up

Running the same test from a client in the same network always givs 600+ down and 200-300 Mbit up

So the down speed dont deliver, but up seems to give acceptable values.

This is tested with both Windows and Linux-vms

first value on the test from now on represent traffic TO a VM and second value FROM the VM

Test from linux client and linux vm

Flexible 910/940 Mbits

Test from linux client and winXP vm

vmxnet3 300/700 Mbits

e1000 190/720 Mbits

Test from linux client and Windows Server 2008 vm (SBS)

e1000 420/890 Mbits

File copy from win7 client (GB NIC) and Win Serv 2008 vm (SBS)

e1000 31/34 MB/s

File copy from XP client (100mb NIC) and Win Serv 2008 vm (SBS)

e1000 4/6 MB/s

I have no idea what to test next to understand were this issue lies. The most strange thing to me is that the speed isn't the same in both directions.

Setup:

Host: HP ProLiant DL380G6 (in HQL)

ESXi 4.0 Build 294855

2x E5530 @ 2,40 GHz

56GB DDR3 RAM

4x Gbit Broadcom Corporation NC382i

1 vSwitch for all VMs (running on several different vlans) with two NICs One active and one unused

1 vSwtich directly connected to iSCSI NAS with 2 nic's with link aggregation (this is working in full speed)

Switch: HP ProCurve 2828

0 Kudos
5 Replies
J1mbo
Virtuoso
Virtuoso

Do you have flow control enabled on the switch? Is the 'raw' performance being measured using (for example) iperf?






http://blog.peacon.co.uk

Please award points to any useful answer.

Unofficial List of USB Passthrough Working Devices

barondavid
Contributor
Contributor

Test between two VMS

XP and Windows Server 2008 (SBS)

e1000(XP) - e1000(2008) 330/190 Mbits

vmxnet3(XP) - e1000(2008) 430/290 Mbits

Note:

Since we upgraded to ESXi the driver on the Windows Servers is:

File C:\Windows\system32\DRIVERS\E1G6023E.sys

Provider Intel Cororation

Version 8.3.2.8

0 Kudos
barondavid
Contributor
Contributor

Hi!

Thanks for a fast answer, no actually Flow Control is NOT enabled, that should really be tested, but i will have to wait until later today since there is some production server running in the host atm.

However, it does not explain that i get some what the same kind of results between two VMs in the same vSwitch and vlan, which should not leave the ESXi out to the switch and back... Right?

And yes Iperf does test raw performance, hard drives are not involved.

0 Kudos
J1mbo
Virtuoso
Virtuoso

Could you give a little more info on the host and VMs..., for example, how many CPU cores does the host have and how many vCPUs are assigned to each guest? Also, for the Linux guests, have you updated the vmxnet driver after installing the vmware tools?






http://blog.peacon.co.uk

Please award points to any useful answer.

Unofficial List of USB Passthrough Working Devices

0 Kudos
barondavid
Contributor
Contributor

The hosts are Windows Small Business Server 2008 64-bit with two vCPUs assigned and 8GB Ram.

The linux VM's run Debian and Ubuntu Server

The XP VM is only set up for test purposes.

On the linux hosts i have just installed the new vmware tools, i thought it handled the drivers during the install... but i may be wrong.

Attached: a schematic of our vSwitch0 were the servers ar plugged in, have tested on different Port Groups, witch same result, but for now ive just tested in the Nissandata-Network

0 Kudos