MPIO inside a guest vm?

[trink408]

Is it supported to configure MPIO inside of a guest OS running on esxi4.1?

I was looking at our Exchange vm configuration, and for some reason the consultant configured our Exchange message store using the Windows Iscsi Iniator. Currently it only has one session to the message store. That said, the virtual nic used for the iscsi iniator is connected to a vswitch configured for Iscsi traffic with 2 adapters, so that should cover the redundacy aspect. Just not sure if it's beneficial to have multiple nics configured inside the guest to use MPIO?

I saw some older posts regarding this, and it sounded like it wasnt' supported, but I thought I would check to see if anything has changed.

Thanks
Kevin

[trink408]

Nobody?

[vmroyale]

Closest you might get is p.68 of the "vSphere Storage" document under the "ESXi iSCSI SAN Restrictions" section:

"You cannot use virtual-machine multipathing software to perform I/O load balancing to a single physical LUN."

Like you said, the virtual nic used for the iscsi iniator is connected to a vswitch  configured for Iscsi traffic with 2 adapters, so that should cover the  redundacy.

[BharatR]

Hi,

Just not sure if it's beneficial to have multiple nics configured inside the guest to use MPIO?

Here is the Link for how to team Two NICS for MPIO in Vsphere

http://www.scribd.com/doc/24586958/Configuring-Left-Hand-ISCSI-and-VSPHERE-MPIO

[trink408]

That link is for configuring MPIO at the host level, not inside the guest. I already have multiple adapters configured on the Iscsi vswitch, and have them set for load balancing/fail over.

I did find one part of that link interesting though, where he spoke about the IO limits of using VMware round robin, and having to change the default values to get the multi-pathing policy to perform better. I'm wondering if those values are the same in newer versions of Vsphere?

[hussainbte]

From my understanding, the only reason why I would think MPIO with guest iSCSI is mentioned as an unsupported configuration is because the multipathing would be totally dependent on the nic teaming and failover policies defined in the vSwitch.

A configuration where you have set IP hash or Load based multipathing in the vSwitch/DVS will obviously load balance the traffic without considering it is meant for iSCSI multipathing.

Issues like intermittent path failures, LUN trashing can then be caused.

If we have to use the Guest MPIO even though we have other alternatives like RDM, vmfs datastores the best approach would be to make the ESXi networking free of multipathing.

1 vmnic using 1 physical nic all the time.

this could in-turn be not cost effective as it becomes difficult for the nics to be used for other traffic(VM traffic)