Comments on a global pool of processors
by Hu Yoshida on Jan 7, 2011
Since my last post on the global pool of processors in the Hitachi Virtual Storage Platform (VSP), there has been a some discussion of this by Nigel Poulton, Scott Lowe, and myself on Scott Lowe’s blog. The Storage Anarchist also threw in a comment. I thought I might summarize that discussion for my blog readers.
It started after Scott read my post on a Global Pool of Processors and speculated in his post that the VMAX and VSP share some architectural similarities in having a pool of processors. Nigel responded and clarified that the processors in the VMAX are tied to either front or backend ports and handle all the I/O processing in addition to other functions like T/F and RDF. These processors’ resources cannot be utilized by other front or back end ports. Nigel went on to say that the VSP had purpose-built processors tied to the front and backend ports to process I/O, but that equivalent functions like T/F and RDF, etc. were handled by a pool of Intel CPUs.
Both Scott and Nigel see value in separating I/O processing from other higher level functionality. However, there was some question about the ROI and time to market of customized ASICs. I responded, and I believe I got agreement, that whether you had customized or commodity components in a storage system did not matter as long as you delivered an efficient, cost competitive, storage system that met customer requirements.
As far as time to market is concerned, Hitachi relies on its R&D capability to deliver the right technology at the right time. We do not have to wait for someone else to develop off the self technology. Often times that means that we deliver technology like switch interconnects, and now separation of I/O and general processors, before the rest of the market.
This discussion was very encouraging to me since some real thought leaders are beginning to grasp the significance of separating I/O processing from the growing amount of general processing that is required to support dynamic tiering, replication, copy on write, and API’s and client/providers that help applications unload SCSI reserve bottlenecks and recover deleted file space. This separation of processing will transform storage systems, from storage containers to storage and data processors.
A global pool of general processors in support of I/O specific processors will be game-changing.
What are your thoughts?
Comments (3 )
I think virtualising workload across processors is a much better approach than hard-wiring them to specific front/back-end ports. I mentioned it in http://www.thestoragearchitect.com/2010/10/19/hitachi-virtual-storage-platform-optimised-architecture/
Happy New Year Chris, thanks for your validation. I read your blog which was right on! Especially your comment that “(this) provides a basis for future enhancements that could be even more compelling.”
[...] of just using storage as containers of data, storage must now become storage computers with a global pool of processors that is separate from the port processors that handle the front and backend I/O processing. This is [...]