Confirming Load Balancing with MD3000i

I don't see mention of an OS, but if you install the Dell software from the resource CD (compatible with Windows 2003 and later, RHEL4 and later, and SLES9 (or 10; I don't remember the oldest currently supported version for SLES) and later), it installs an MPIO driver. This driver would split the IO on an iSCSI level between NICs to be able to give you load balancing. This does assume you have the cabling set up like in the cabling diagram on this Dell post.

2008 64bit. currently the server is connected to a backup switch where both ports of the md3000i also connect. the md3000 only connects to a single switch, no need for high redundancy in our scenario. i would think this should be irrelevant. i would expect that both controllers would still be seen and that if the speed is not aggregated that at least multiple sessions should each use a different controller and hence route through a different nic. also of note, both controllers are using the same subnet since its a single switch and same backup network ip range. again, i cant see how this matter in this case and its not splitting switches and hence should not be cause for duplicate ip's from the mpio routing algorithm.

As it's not working, and all documentation for all iSCSI SANs that I've worked with and read up on, always say not to team NICs, I would suggest you follow best practices and see how that works out for you.

Responses in-line... kind of :)

-Mohan

I would like to confirm this since I have not been able to find a definitive answer. It may be a little long winded so I apologize in advance.

I have an md3000i that is/will connect to two additional md1000 enclosures for our backup environment. The md3000i will connect to a dedicated backup switch. I have the backup server using two 1gbps nic's aggregated for a total of 2gbps. Now, its my understanding you cannot aggregate the ports from the md3000i. So what Im trying to achieve and expecting to see is that the 2gpbs of the backup server should be used, by splitting the volumes across the different controllers and ports on the md3000i. I would expect to see one path used for one volume and a second path used for a volume on the second controller - this is not happening.

NIC teaming is not needed to aggregate bandwith. However, for maximum performance, Dell recommends having 2 virtual disks (on 2 separate disk groups if possible), each owned by a separate controller. The MPIO driver will ensure that all "available" paths are used.

When I created and mapped the first volume i can hit a consistent 50% of the combined aggregated bandwidth from the backup server. Monitoring traffic stats its using on NIC. When I created a second volume on the second controller and perform a transfer its still only using one nic. Fair enough. So I intiated two transfers simultaneously, one to vol1 on controller 0 and one to vol2 on controller 1. It still only used one link.

It appears to me that the iSCSI sessions to both the controllers have the same source port. That would be the reason why you have only 1 link being used. My recommendation would be to setup 4 iSCSI sessions when you are using 2 ports total on the MD3000i. 1 session from each of the source ports on your host to each of the controllers.

In my iscsi initiator i had the IP for each of the two controllers listed as the portal. I am using (only using 2 of the 4 for testing). Before I had the mpio drivers installed, I could access only one volume. Although the second volume would show in disk manager i was unable to initialize or bring it online. So I installed the mpio drivers. I now have access to the second volume but am receiving these results.

The second Virtual Disk that you were seeing was basically the same VD seen from the second path. Thats why you could not initialize it. The MPIO drivers work to show a single disk to you.

So since the md3000i will not use port aggregation (I believe) i would like the ability to get max performance from the unit and i see that only happening in the way I mentioned above. Create different volumes on each controller / path so that each volume/path should traverse a different nic in the bonded team.  This is presently on a 2960g but will be upgraded in the next few weeks to a 3750. Im using standard lacp mode (not etherchannel). Would this have anything to do with mac vs. ip hash or source / destination?

We also have a white paper for tuning performance. One thing to note is that performance depends on a lot of variables and you might not be seeing line speed for transfers:)

http://www.dell.com/downloads/global/products/pvaul/en/pvault_md3000_whitepaper.pdf

Am I looking at this wrong? Do I need to scrap the nic teaming and rely on mpio solely for these results? If I use mpio what is the proper config for that scenario.

All supported configs are listed in the MD3000i Support Matrix : http://support.dell.com/support/edocs/systems/md3000i/en/index.htm

Thanks