Network Load Balancing

[six4rm]

We have 12 IBM Blade servers and we recently noticed that outbound network traffic from each ESX isn't being load balanced. The vSwitch has two physical NICs assigned to it but traffic is only being transmitted on one of them . NIC Teaming is configured for the vSwitch and is set to "Route based on ip hash".

Update 5 is installed on each server.

Any ideas?

Thanks in advance.

[AWo]

Are both NIC's are listed under "active" in the vSwitch configuration?

Change the policy to "Source Port ID" and check if the load is balanced then. if so it is due to the nature of the actual communications taking place that one adapter is not busy (yet).

AWo

\[:o]===\[o:]

=Would you like to have this posting as a ringtone on your cell phone?=

=Send "Posting" to 911 for only $999999,99!=

[weinstein5]

Even though it is called load balancing it is not technically load balancing - with IP hash the vmkernel examines every packet and based on source and destination IP address a physical NIC will be selected for that packet to go out - so if the majority of your traffic is from one vm to one destination all that traffic will go out one NIC - Port Based which is the default the NIC is selected based on the virtual port the virtual NIC is connected so even in this case if a single VM is responsible for the majority of traffic only one physical NIC will be used -

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful

[six4rm]

Thanks for the replies.

The specific server that the screenshot relates to has 24 live VMs. Our users mainly use one web based application, which is hosted on a physical server, so the majority of network traffic will have the same destination. I would still expect to see traffic balanced across both physical NICs. Both NICs are showing as active within the vSwitch configuration.

It was only this morning that I changed all 12 ESX hosts to "Route based on ip hash" rather than by Port ID in an attempt to balance the network traffic. It's still as unbalanced now as it was before.

[AWo]

So by using the port ID algorithm you still have this single NIC "underutilized"?

AWo

\[:o]===\[o:]

=Would you like to have this posting as a ringtone on your cell phone?=

=Send "Posting" to 911 for only $999999,99!=

[depping]

You should only be using "IP HASH" when you have also "teamed" your ports on the physical switch, in other words created a channel. If you haven't than Virtual Port ID is your best pick.

There's a know issue by the way and a work around to this issue. You might want to check the following KB:

http://kb.vmware.com/kb/1013691

Duncan

VMware Communities User Moderator | VCP | VCDX

-

---

Now available: Paper - vSphere 4.0 Quick Start Guide (via amazon.com) | PDF (via lulu.com)

Blogging: http://www.yellow-bricks.com | Twitter:

[six4rm]

Yep, unfortunately so.

It's currently set to "Route based on ip hash" and the screenshot attached is the current performance data. I've just changed it to "Route based on the originating virtual port ID" and I'll post another screenshot in 30 minutes or so.

[six4rm]

Hi depping,

On the physical networking side of things each physical NIC within the ESX hosts connects to a seperate physical Cisco switch within the IBM Chassis. These two switches then connect to another physical Cisco switch via a four port EtherChannel (one per switch) which all our other servers are also connected to. I hope that makes sense...

So, am I not actually gaining anything by setting the vSwitch config to "Route based on IP hash"? I'd still expect to see outbound traffic balanced across both physical NICs though right?

[six4rm]

Here's the latest performance stats after changing one of the ESX hosts back to route based on originating port ID.

I've read the KB that depping posted, however does it not only apply to build 64607? We're running 213532 across all 12 hosts. I also cannot see what checkbox that point 8 of the KB is relating to.

[depping · Accepted solution]

1) IP Hash should be avoided in your setup.

2) According to the KB it does not apply, but just try it. Seen it once already that this work around did work in a later build

Duncan

VMware Communities User Moderator | VCP | VCDX

-

---

Now available: Paper - vSphere 4.0 Quick Start Guide (via amazon.com) | PDF (via lulu.com)

Blogging: http://www.yellow-bricks.com | Twitter:

[six4rm]

OK cool. I've changed the global vSwitch0 config back to route based on originating virtual port ID. I've also just changed the Production LAN config as recommended in the KB. Screenshot attached.

Looking good!