So what about the VSP G1000?
After my rants in my previous two posts about HDS software commonality and adoption of industry standard hardware, I wanted to broach this particular question for a number of reasons. My rant about commodity processors being good ‘enough’ (and yes that is the term I used) for the midrange market, and the reality of the changing dynamics of the storage market becoming midrange heavy, will no doubt raise questions on why we still need high-end or frame based arrays.
My response to this always comes down to either a single simple comment: “it's about fitness for purpose” or a single question: “What are the use cases?” In fact, HDS as part of its launch, talked about use cases with the rhetoric around “software defined, application led” – with “application led” acknowledging that for some applications pure commodity is fine (think scale out Hadoop clusters maybe) while others need more traditional infrastructure (think large, centralized OLTP databases with high data protection requirements). This application led discussion is nothing new but there is something I do want to address here. Large web-scale organisations, who have a global user base (ie, Google, Yahoo, Facebook, ebay etc.), have largely changed the popular psyche around how applications should be behave by pushing the 3rd platform concept. I speak to customers daily that want 3rd platform like applications (which is a positive) with little to no understanding on the technical efforts involved in re-platforming (because almost all of them are managers, not developers), let alone the organisational challenges around process and change management (which is a negative). The reality of these challenges effectively means that a rapid shift isn’t going into occur any time soon and a lot of applications will continue using 2nd platform like characteristics for a long time to come. To create some context on this effect, you don't have to look hard to see that mainframe is alive and well in the world (some define this as 1st platform) despite the fact that Open systems (2nd platform) have been around for 20+ years and we are already seeing 3rd platform architectures (and even platform 2.5, which I’ll address in a later post).
Realistically (and my friends at EMC would agree), large frame based arrays are still required to meet specific use cases around RAS (Reliability, Availability and Serviceability). These are key driving characteristics for VMAX over VNX or XtremeIO, and the situation is no different here with HDS. Large frame arrays, with their multi-layered architectures, are designed to be extremely fault tolerant and do not degrade (performance) heavily during failures. They're designed to offer complex replication technologies, and are designed to deal with a large number of diverse workloads. While the Gx00s running SVOS are fantastic arrays, they do not have the same highly layered architecture and the reality is that there is nothing you can do with commodity hardware in a flat two controller architecture to achieve the same level of RAS. This is because such high levels of RAS are a by-product of the architecture itself, not the components. EMC proved this when they built the VMAX3 using commodity hardware. Does this mean we will see a frame based array from HDS that just uses Intel-multicore? Your guess is as good as mine (well probably not), but only time will tell.
There is some argument that high levels of RAS can be achieved through scale-out architectures when combined with a 3rd platform decoupled applications architectures on a case by case basis - which of course I am in favour of, as is HDS. However, as I ranted about above, a complete shift to these application classes isn’t going to happen anytime soon. It is pretty clear we’re at least a generation or two away from re-platforming occurring on a scale large enough to remove the need for such systems to exist. That’s not even considering the effect of the requirements around platform 2.5 will have if/when it starts to gain serious traction. Even assuming it won’t or other architectures can meet the requirements just as effectively, the market for such systems may shrink (much like the mainframe market), but it's not going away any time soon. Until that time comes, it is good to see that Hitachi is taking a leadership role and still focusing on providing innovation and solutions (much like IBM still is in the mainframe space decades later and still selling them like hotcakes! Mmm hotcakes).
That's it for this series! Big thanks to rmadaio for being a sounding board for much of my ranting that went into this!
I already have a few non-Gx00 posts going which I'll hopefully get around into next week.