iSCSI HBA redundancy using Accelerated iSCSI Feature Pack

[Christopher Barlow]

We are setting up a redundant server database storage platform for a customer. We are running a DL360 G5 server with an integrated dual-port NC373i NIC. The SAN is an MSA1510 with dual controllers, connected to two switches for fault-tolerance.

The idea is that we have a fully redundant network path, such that if we lose a server NIC port, switch or RAID controller, the system will continue to run.

The problem I’m having is a configuration issue, with how to properly achieve NIC port redundancy when using the NC373 NICs as block-level iSCSI HBAs, in particular for TCP Offload (using Accelerated iSCSI licenses).

If I use the SOFTWARE iSCSI initiator, I am able to achieve the goal by simply using an ethernet NIC team, and specifying an IP address for the logical teamed adapter, and selecting this IP when setting up the targets in the Microsoft iSCSI Initiator utility.

How do I achieve redundancy using the NC373s as block-level iSCSI HBAs once we've installed the Accelerated iSCSI licenses?

As far as I can see, the iSCSI HBAs are completely separate interfaces from the ethernet interfaces and from the logical team we created earlier. When you go to set up the targets, we able to select the iSCSI HBAs, which are using the individual iSCSI IP addresses (on the same subnet), for each adapter we defined in the NCU – but we can only select one or the other.

I cannot figure out how to bind to a target using the Initiator software using more than one HBA, and in doing so achieving fault tolerance. I do not need load balancing, active-passive failover would be fine. I have tested in single HBA config and done a comparison of CPU activity against using the software initiator, and there is definately a much reduced CPU impact during file transfer to the SAN. Now I need redundancy.

To me it looks like the this is a configuration issue with the Microsoft iSCSI Initiator programme. But I may be wrong. Is there a recommended way to configure the initiator, target portals and iSCSI IPs to achieve this? Have I got completely the wrong end of the stick?

Thanks for reading!

[HEM_2]

You are correct that NIC Teaming does not provided accelerated iSCSI HBA redundancy. The HP multifunction NICs use a separate stack for iSCSI altogether.

I think you might want to look into Microsoft MPIO as the redundancy solution but I'm not sure how all the pieces fit together.

[Sean van Osnabrugge]

Just a note, Network Teaming is not supported by Microsoft on the iSCSI interface: http://www.microsoft.com/WindowsServer2003/technologies/storage/iscsi/iscsicluster.mspx

MPIO or MCS is the way to go to provide fault-tolerance.