It seems that in this high technology industry of ours, it is fashionable to take sides or belong to a clique. One has to be an open source supporter because “Big Software” companies are evil; one should only pray to the Hadoop deity as any other framework is passé; and heaven forbid one is not a card-carrying member of the Software Defined Everything party. I personally think that such behaviors are small minded and in total ignorance of the harsh reality of IT life.
As I return from Open Compute Summit V and at the risk of being called a Judas by my colleagues of the storage industry, I want to use this blog to applaud the Open Compute Project “open design storage” efforts; but because I don’t want to be in a clique, I want to declare that purpose-built storage (I prefer this word to “proprietary”) has much value and will be around for a while.
The Open Compute Project was born of the desire by a consortium of web properties to put the power of infrastructure design in the hand of those who purchase the most and, thus, reduce said infrastructure cost for hyper-scale deployments. Initially focused on server architecture and progressively moving to the rack, storage and network space, OCP is creating open reference designs which could then be implemented and innovated upon by ODMs and OEMs. The common designs drive supply chain optimization and incremental innovations, and they foster competitive differentiation within an interoperability framework.
DDN has supported OCP since early on – the core tenets align perfectly with those of our WOS Cloud Storage product line. Our rationale is that OCP storage designs ultimately help us streamline our hardware supply chain, reduce our costs, allow our R&D investments to be focused on software which is what makes WOS automagically superior!
Back in 2011, it was obvious that the overall process would need time to bear fruits and mature. And here we are in 2014 and finally it’s all a reality! Yesterday, we were happy to announce that WOS is now available on OCP certified (and inspired) hardware solutions through our partner Hyve Solutions.
Hyve Solutions data center rack infrastructure is a highly efficient open architecture designed to reduce the number of data center components while lowering TCO in the scale compute space. DDN WOS is high-performance object storage capable of storing up to 983 petabytes and 32 trillion unique objects, delivering hundreds of millions of objects per second, at the lowest TCO. This is a joint step in the right direction in DDN’s commitment to work with infrastructure partners to build next-gen storage solutions and achieve scalability breakthroughs to support large cloud and webscale computing applications. The DDN team wrote more on the subject.
For this I say: Hail to Open Compute Storage!
This being said, we does not only sell WOS Cloud Storage. Many of you readers know, we earned our stripes in the High Performance Computing market – today we are supplying purpose-built high performance storage to 8 of the top 10 supercomputers in the world. HPC is a very demanding workload where manufacturers are challenged to solve problems that look insolvable at first. In fact, the nature of the HPC business is such that the customers often order systems that do not exist at the time of placing the PO and the supplier commits to deliver them when they are just a glimmer in some engineer’s eye. There is no “open design” that can fit the bill and systems are “proprietary” or “purpose-built” to satisfy the grueling requirements of this market. This is not because the bright minds of the OCP community could not do it, but because the business justification is not there.
As a general rule, storage systems built for the HPC market (and tangential markets like the Intelligence community or the Media and Entertainment space) have very rigorous requirements in term of QoS, throughput, latency and density and cost – to name a few. These requirements drive designers of such systems to be very, very tight on margin of errors and very resistant to drive failures. It’s not uncommon for such systems to have 10,000 to 30,000 drives all spinning and delivering data at the same time to tens of thousands of cores. Let’s not forget that a millisecond lost per operation on a billion operation workload equates about 11.5 days lost! Without going into more gory and geeky details, this is why there is still a need for purpose-built systems and there will be for a long time. These are critical to the IT infrastructure just like OCP based systems are. And, by the way, purpose-built does not mean proprietary silicon or black magic. Our SFA products are purpose-built but are made of 1.5M lines of code running on x86 based open system component.
For this I say: Hail to purpose built storage!
I miss the intimate collegial atmosphere of the early days of OCP where we could all fit in a small restaurant but I applaud the amazing success of the organization as witnessed by the hundreds of attendees in San Jose last week. See you next year at Open Compute Summit VI!