Yesterday my friend Enrico Signoretti shared an article about Microsoft open sourcing their next-gen cloud hardware design. This and another post from my friend Tom Hollingsworth on the Facebook Wedge 100 TOR switch and finally some mixed thoughts about a specific company presenting at Storage Field Day 11 had me pondering about open sourced hardware designs.
Get ready for another sci-fi fantasy post out of my tortured mind.
Open Source Hardware Design – Why?
First of all, I would like to exclude from the scope of this article the question around free hardware design vs open source hardware design and the many intricacies of various licensing models (GPL, OHL etc).
The idea behind open source hardware designs is to share one’s experience with building hardware systems from readily available components on the market in order to allow others to build similar systems, improve, adjust, and modify them as needed. Is that a purely philantropic purpose? Yes and no.
One potential reason for open sourcing hardware designs is to present research in future concepts to the larger data center community, seek validation, comments / propositions for improvements, and factor in these recommendations into later iterations of the design. This way, the company or consortium releasing an open sourced design benefits from free advice, and so does the larger community assembled around a given open sourced hardware design.
On the data center playground, the initiators are mainly enterprises with a large DC footprint such as Facebook and Microsoft, to name two of them. Their unique level of operation at scale means that every little inefficiency is multiplied in the tenths of thousands, whether it is power consumption, space usage within a chassis, thermal dissipation or the very way the hardware architecture operates. Mainstream vendors are not always able to come up with a satisfying solution, furthermore as more and more features and requirements are moved up to the software stack (think “built to fail” apps, with high-availability or tolerance to failure features embedded at the software level), the hardware can either be thinned down to the bare requirements of compute resources, or beefed up to fulfill a given purpose.
A killer for innovation?
The devil’s advocate would say that open source designs are prone to kill investments in R&D, to cut down innovation and reduce the generation of intellectual property, which in turns means a net loss for hardware manufacturers.
Reality check: we are already in a market that has been largely commoditized, where components are similar and readily available. If we focus our attention exclusively on x86 server hardware, most systems share the same characteristics: same CPU architecture, same chipsets from the same vendors, same memory modules, same storage devices, all that we see is a variety of form factors (1U, 2U, 4U) and different disposition within the chassis, maybe more RAM slots here, less NIC ports there.
It is a market where profits are low, where the competition fierce, and where vendors agressive in their discounting policy. People working in the channel know well that you won’t make tons of money if you’re just swapping pizza boxes here and there.
Let’s also not forget that there will still be niche use cases where a vendor may want to develop their very own hardware architecture to address specific demands (hello Flashblade) and let’s also consider that, even in the x86 world, there are specific architectures (specifically blade systems) that may be preferred to commodity rackmount servers.
Open sourced hardware designs might also play a role with software-defined storage solutions. Just like the Intel Storage Builders initiative focuses on bringing innovation to the storage world by leveraging Intel technologies and research, one could very well envision a consortium of SDS companies & startups contributing to achieve a standard, open-sourced hardware design that matches their requirements. True, x86 commodity hardware is good enough. But surely there are tweaks that could be made?
Max’s Opinion
What I am saying is potentially an ineducated man’s utopia. Maybe hardware vendors would gain in participating in open sourced hardware designs. Why? By sharing open sourced designs with competitors, this would help them overall save on R&D costs (by pooling them across participants) while fostering innovation. The R&D savings on hardware design could be injected elsewhere, maybe moving them a level upper at the software layer, or in designing new, revolutionary hardware components.
Such open sourced hardware designs might help standardizing the data center offerings. It could also cut down production costs and in the end the customer could be the winner. To answer the question in this post title, maybe open sourced hardware designs are not a next step beyond commoditization, but only the latest or future iteration of commoditization.
The innovation that is happening in the large data centers of the Facebook, Google, Microsoft, Rackspace and the likes will eventually be cascaded down to the larger market, and we might see such offerings become the standard in the years to come.