I met with our EMC account team last night over dinner to discuss a number of items, including optimal stripe depth configuration, migration techniques for moving between different stripe depth configurations, iSCSI, and the EMC Technology Summit. It was a very enlightening conversation. The majority of the conversation was centered on cylinders and stripe sets. For those new or not familiar with EMC jargon, a cylinder in their world is equivalent to a disk platter, not a spindle as one would intuitively expect. There are a number of EMC customers in the world that have maintained a four cylinder stripe depth configuration rather than the default two cylinder stripe depth configuration. Reasons for operating a four cylinder stripe depth traditionally center on specific application performance. Some application vendors believed at some time in the past that four cylinder stripe depths were more efficient than two cylinder stripe depths based on their own performance analysis. EMC agrees that this may have been the case with Symmetrix technology prior to Sym 4.8 generations of their product. I’m hard pressed to find people in the field that have implemented a Sym 4.8. In any case, it boils down to this: accessing four cylinders serially on the same drive (four cylinder stripe depth) is less efficient than accessing four cylinders on two drives in parallel (two cylinder stripe depth). EMC has done extensive performance analysis on this comparison which they have summarized in two very succinct slides.
Migrating from a four cylinder stripe depth to a two cylinder stripe depth is pretty straight forward. It is a host level migration. You add disk to your Sym. (Gee. Who would have thought?) Configure equivalent size volumes to your production volumes using a two cylinder stripe depth. Then, mount those volumes on your hosts and move your data between the volumes. This is the general approach. As always, the devil is in the details. If you are moving flat files and documents, this approach can be taken pretty much at face value. If you are moving database volumes, things can get pretty sticky based on your specific RDBMS. For Microsoft SQL Server, you have to shutdown your servers, move the database files, delete the old volumes, rename the volumes to be the old drive names, reboot your servers, and restart your SQL Servers. Giving a little prayer during the reboot wouldn’t hurt. As you may have guessed already, there is downtime associated with type of migration. The downtime associated with this migration needs to be weighed in the context of the impact on your business.
We then discussed iSCSI. There wasn’t much to discuss. EMC is still working on their iSCSI story and products. You can readily see the adoption, or lack of, in a search for iSCSI on the EMC web site. My interest in iSCSI is in obtaining a lower cost storage solution with performance some where between a Fibre Channel SAN and NFS NAS. EMC will be making announcements around iSCSI at the EMC Technology Summit at the end of this month.
The EMC Technology Summit is going to be a blast. I have signed up for the iSCSI adoption breakout session and the session on providing on demand storage solutions. I am looking forward to four days of high intensity knowledge acquisition.
