Techno Musings

As a part of ongoing architecture discussions, the concept of commodity has come up.  First I would like to start with a few questions on the point.
< --more-->

• Does commodity mean only Intel based architectures?
• Does commodity mean Linux or Windows based operating systems only?
• Does commodity mean hard disk drive technologies only?
• Does commodity mean systems that are produced in bulk by Taiwanese or Chinese manufacturing?

Well quite simply the definition that I found on Wolfram Alpha for commodity is: articles of commerce.  That is a pretty simple definition, but it doesn’t meet I think what my competitors want to call commodity, note that is just a guess.

To better understand things, I’d like to look at IBM for a minute.  IBM uses their POWER architecture in the pSeries, iSeries, the DS, and in BlueGene.  Since IBM produces all of these products and uses them, and also manufactures POWER processors for others does that mean that they are or are not using commodity processor elements?  What I surmise that IBM gets out of reusing these components within their systems is an increased component volume that provides a cost efficiency.  Hmm, it gets you thinking doesn’t it.  However, IBM does recognize that there are reasons to not use commodity components when there is more benefit than cost.  More specifically look at BlueGene which uses custom compute and link ASICs to push the envelope.

Now I also want to talk about HNAS for a minute in the same light.  HNAS uses well FPGAs produced by Altera.  FPGAs, just like AMD processors, Intel processors and IBM processors follow Moore’s Law in terms of the number of gates increasing over time.  Further HNAS does not use modified special purpose FPGAs from Altera, but instead utilizes off the shelf FPGAs to get  efficiencies because Altera mass fabricates  FPGAs in bulk for a mass market — note that Xylinx competes here too.  Further the HNAS system makes use of mass market ASICs and CPUs where possible for things like FCP.  In fact there isn’t a component in the HNAS system that is not somehow tied to a mass market component.  The magic comes from, well the software on the system and the fact that it can be accelerated by FPGAs.

Next onto the USP-V and the AMS, which both include ASICs for Hitachi’s RAID algorithms.  This is an intentional design choice as Hitachi wants to ensure that our RAID implementation is extremely robust.  Actually it is really simple the choice of putting RAID algorithms removes many testing patterns and corner cases because well it is hardware.  This alone means that when you compare an ASIC (and an FPGA based approach too) to a 100% software approach it will be inherently more reliable. Further like EMC’s link ASIC on their new V-Max the USP-V also uses ASICs as a part of our HiStar switch.  We also make use of mass market technologies within the system where appropriate to take advantage of cost infrastructures.

The point that I’m trying to throughout this post is that all of the storage and systems providers make use of components to improve the cost efficiencies of their solutions.  Where we see an advantage in using hybrid systems that include FPGAs or ASICs we do so.  Hitachi’s bet is on embedding our RAID algorithms and in our switching infrastructure, like EMC.  It is our unique value that our customers believe in as it improves both scalability, how green our package is and in reliability.

So what was the purpose about arguing that your systems use commodity components again?