I was reading Hu Yoshida's Blog over at HDS where he discusses both server virtualization and storage virtualization. In the blog, Hu makes some interesting claims about the evolution of storage virtualization (at the block level). Hu defines three phases of storage virtualization:
- Storage Virtualization 1.0 - appliance-based approaches that began to appear around the year 2000. According to Hu, all these "1.0" approaches did was offload volume management from the host. In reality they did (and do) far more than that with the ability to migrate data without interrupting host access, thin provision volumes, and provide hardware independent snapshots being just a sample of the capabilities.
- Storage Virtualization 2.0 - based on Hu's definition, this is the HDS Universal Storage Platform. Apparently the big things about storage virtualization 2.0 are "volume mobility", iSCSI/ESCON/FICON support, and use of gateways to provide other services such as CAS and NAS. Of course (with the exception of ESCON/FICON support) all that was possible with many of the 1.0 appliance-based solutions, years ago! Hu then goes on to make claims for performance and scalability vs. appliance-based solutions for which there is precious little evidence. If you look at the one source of independent storage benchmarking (www.storageperformance.org) it makes interesting reading. There are two results that are particularly enlightening:
- IBM's SAN Volume Controller ( a solution built on commodity x86 hardware) with a score of 272,505.19 SPC-1 IOPs at $12.05/IOP for 24 TB of usable storage, published in July 2007
- HDS Universal Storage Platform with a score of 200,245.73 SPC-1 IOPs at $17.61/IOP for 26 TB of usable storage, published in October 2007
- Storage virtualization 3.0 – this is where we get such advanced features as thin provisioning found in the Universal Storage Platform. But that ignores the fact that thin provisioning has been available in the “1.0” platforms for the best part of half a decade!
So, what we have is definitions for the three ages of storage virtualization which don't seem to have much bearing on reality except as defined by HDS product marketing. Don't get me wrong, I think USP is very good array with some powerful features, but this sort of self-serving redefinition of storage virtualization only further muddies some very dark waters.
Posted by: Nik Simpson