Best practice to configure management network redundancy in ESXi

[rshroff]

Hi,

Pardon my ignorance if this has already been answered previously and be kind to point me to the relevant link/forum.

I was interested in knowing the best practice while configuring management network in ESXi. If I team two pNICs, the idea is to put both the adapters to work thereby ensuring failover and load balancing. However, I understand that true load balancing in case of single vNIC VMs occurs only if the load balancing algorith is set to "ip hash". Is this the case with kernel NICs as well?

Also, is it recommended to have two VMkernel ports on two vSwitches (each with it's own pNIC and IP obviously) or to have a single vSwitch with a single kernel port but two teamed pNICs? At least I save on an IP in the latter scenario and my DNS looks much cleaner as well. However, I would like to hear from the experts.

Awaiting response...

Thanks in anticipation.

Rolf

[djciaro]

Nice post on yellowbricks about this issue worth a read:

http://www.yellow-bricks.com/2011/03/22/esxi-management-network-resiliency/

[rshroff]

Appreciate your prompt response djciaro; the article was very helpful. Alluding to the last post on that article, has the scenario changed with ESXi 5.0 as I'm using the latter. Hope the best practice in the article still applies to ESXi 5.0.

Rolf

[djciaro]

Hi Rolf,

Sure do, Best practices still stand for ESXi 5.0

Best of luck

2011/11/26, rshroff <[e-mail]>:

rshroff replied to the

discussion

"Best practice to configure management network redundancy in ESXi"

To view the discussion, visit:

http://communities.vmware.com/message/1870223#1870223

[rshroff]

Thanks djciaro.Just one more thing...

I'm sure there would be a scenario when a 1Gbps pNIC may not suffice for vMotion traffic and we can scaleout using multiple pNICs for the vMotion port group. My question is, will both the pNICs be used equally in a true load balancing fashion? If not, what does it demand to achieve true load balancing in case of VMkernel ports?

Rolf

[rickardnobel]

rshroff wrote:

I'm sure there would be a scenario when a 1Gbps pNIC may not suffice for vMotion traffic and we can scaleout using multiple pNICs for the vMotion port group.

In 4.x there are only one pNIC actually used for vMotion no matter which load balancing option being used. In 5.0 we have multi-NIC vMotion where several pNICs could be used together during a single vMotion.

rshroff wrote:

My question is, will both the pNICs be used equally in a true load balancing fashion? If not, what does it demand to achieve true load balancing in case of VMkernel ports?

Since the VMkernel ports could be used for many things (management, iSCSI, vMotion, NFS, Fault Tolerance) there is different configurations that has to be done have any load balancing. Which of these use cases do you want more "real" load balancing over pNICs?

[rshroff]

Thanks Rickard...my response inline...

rshroff wrote:

I'm sure there would be a scenario when a 1Gbps pNIC may not suffice for vMotion traffic and we can scaleout using multiple pNICs for the vMotion port group.

In 4.x there are only one pNIC actually used for vMotion no matter which load balancing option being used. In 5.0 we have multi-NIC vMotion where several pNICs could be used together during a single vMotion.

So I take ESXi 5.0 is smart enough to do true load balancing for vMotion based traffic with no concern to the load balancing algorithm specified for the corresponding port group.

rshroff wrote:

My question is, will both the pNICs be used equally in a true load balancing fashion? If not, what does it demand to achieve true load balancing in case of VMkernel ports?

Since the VMkernel ports could be used for many things (management, iSCSI, vMotion, NFS, Fault Tolerance) there is different configurations that has to be done have any load balancing. Which of these use cases do you want more "real" load balancing over pNICs?

I was referring to the case of VMkernel ports used for vMotion traffic. Actually, if I go by the setup based on the yellowbrick article (posted on this thread earlier), I can use say 3 pNICs, one for the Management port group (with 2 standby pNICs) and other two for the vMotion port group (with 1 standby pNIC used by the management port group). In such a configuration scenario, I would expect ESXi 5.0 to do true load balancing when it comes to vMotion traffic. Please correct me if I'm wrong in my understanding.

Rolf

[rickardnobel]

rshroff wrote:

So I take ESXi 5.0 is smart enough to do true load balancing for vMotion based traffic with no concern to the load balancing algorithm specified for the corresponding port group.

ESXi 5.0 is smart to do it, if it has been setup in a certain way, described for example on Yellowbrick, so it has to be done with Port ID load balancing and Active/Standby configuration, several Vmkernel ports with IP and so on. If this has been setup then the vMotion traffic is load balanced (in a more "true" way actually than other VM traffic) across the available VMNICs.