Link aggregation and MPIO

[x-cimo]

I am trying to figure out the most efficient network for sharing the NAS to a ESX server (iSCSI) and other host on the network.

However I am not sure if this would work as I intend it to work, or it's overkill / wrong.

The goal is to have max speed to host or to the ESX server.

It seem that link aggregation is only useful if there are multiple clients, as a single client cannot be balanced over link aggregation.

MPIO is therefore the only way to balance iSCSI traffic comming from the ESX.

I am however, not sure if once the iSCSI MPIO traffic get on my switch if it would get properly balanced over the aggregated link to my NAS? I think it would because it would be comming from different NIC on the ESX, but can anyone confirm that?

Here is an image of what I plan to build.

[eisi]

Hi x-cimo

Basically I had the same problem - but my box is a RS3411RPxs (10 bays, 4x 1GBit/s). I'm also using ESX to connect to our storage box. Due the fact that link aggregation is not suitable in this situation I implemented Multipath I/O - means in your situation that you have to define two seperate networks, say one belonging to your official LAN and a complete different one, which is only used by your esx host and your storage box. Then you add in esx5i your software iscsi hba. On your storage box you have to setup an iSCSI target which allows multiple connections. Thereafter you discover your Storage box from the esx. After then you have to adjust your iscsi paths on your esx so it uses round robin access to the i/o paths (standard is to use only one link and in case of a link failure it will switch to the other link).

cheers,
rOger

PS: you could not use link aggregation from the switch to your storage box. That is a small drawback only if you have many clients in your official LAN...