Techno Musings

A huge portion of Hitachi’s approach to storage is all about ensuring data safety.  At the core of our storage culture we very much recognize that you as our customers are entrusting us, Hitachi, to be the careful custodians of one of your most valuable asset: your data.  Data and its sister Information (note information is data in a context) are interesting things which have fascinating properties like having their value go from worthless to valiant savior of the company — just think of finding that one email out of the haystack proving your company is correct in a court of law.  Protecting critical information assets, note my change in terminology here, is a big deal for Hitachi.  You can see the incredible care that our company takes in relatively simple things such as the distribution of where our engineers live relative to their office.  In fact a super majority live within just a 10-15 minute train ride to the nearest station here in Japan. Or in the fundamentals of our technology choices in our enterprise platform: ASICs for critical components versus software-only implementations.

Let’s face it top companies in the world deploy hybrid computing platforms for a variety of tasks.  In the past I’ve talked about this point in several posts such as “The Dance“. Example companies include IBM via BlueGene/P is pushing the envelope through the usage of compute and internode communications ASICs, companies like CISCO and Brocade with high I/O (network in this case) demands make use of ASICs and FPGAs, companies like 3PAR and Hitachi deploy ASICs in storage applications.  While there is NRE cost in building an ASIC the key is to find out where the cost benefit curve is, where your organizational skill set is, and then pick the most appropriate design approach.  There is a great article on CISCO’s usage of ASICs over at EE Times Asia, and from that article I want to pick up one quote in particular from a Gartner representative about CISCO’s approach.

‘”Cisco is doing fewer ASIC designs than they have done in the past, but the revenue per ASIC design is growing, causing them to be one of the top buyers in total ASICs and the top buyer in wired comms,” said Lewis. “In other words, Cisco is very selective in what ASICs they take on each year, but each design they do internally has been very successful in generating significant revenue for them,” he added.’

This illustrates my point exactly where a company feels that they can generate differentiation they will spend R&D dollars there and ASICs are no different.  Furthermore , leading edge R&D from Microsoft Research called the Barrelfish Project is working towards an approach whereby applications tell an operating system the kinds of services they require and the OS responds accordingly.  A fundamental tenant of this OS are many heterogenous cores or many-cores within a single system, where FPGAs, IBM’s Cell Processor, and GPUs are all cited as these different types resident within a given system.  So even Microsoft, de facto king of software, realizes that hybrid computing is only becoming more and not less prevalent.  Okay so one of the reasons I spent my time on this was to prove the point that we are in a time of hybrid computing, and not a time where software only approaches are 100% the norm.  Even those that claim a software only approach is the only place for innovation still use ASICs in their platforms, still make use of Application Specific Standard Products or NICs, or HBAs, or, or, or.  Don’t be fooled by the rhetoric.

Now that the groundwork has been laid, here’s the argument for my case: Hitachi views the fundamental building blocks of data protection, the RAID algorithms, as so important that we’ve encoded large portions of them onto ASICs  This is something that as a our customer should help you sleep at night.  The reason is that it is hardware and is therefore less flexible, which in the case of data protection algorithms is a good thing.  You see with a predominately software approach there are infinitely possible test cases, corner cases, etc. from a strict QA perspective.  While an ASIC (or Application Specific Standard Product) with inherent non-reprogramabilty will have a finite number of test cases, corner cases, etc.  That is well because it is hardware and not software, as I mentioned before.  Here is an example using a very common integrated circuit, the 555 timer.  A 555 timer is a very simple IC that acts like one and only one thing an “IC Time Machine.”  You cannot run C or Java code on the 555 timer, instead it is an Application Specific Standard Product having a deterministically defined set of pins and functionality.  Basically a system applies power to the IC and it behaves in just one way for actual use in production or during a quality assurance test.  Similarly our approach to RAID, to encoded in ASICs, also has deterministic behaviors.  It is only going to act in one way when the power is applied to the chip.

So there are a couple of questions that I want to leave you with.  With a software only approach to RAID, how does one expect to magically eliminate all of infinite test cases?  Second with the industry attempting to move towards modular architectures and including intentional price premiums but having modular cost structures, who is benefiting from these new systems: the user or the vendor?

Personally having someone who views themselves as a custodian of my data, and spends R&D dollars going that extra mile to pick the right place for differentiation sounds better to me.