In this guest post, James Bailey, director of datacentre hardware provider Hyperscale IT, busts some enterprise-held myths about the Open Compute Project
Market watcher Gartner predicts the overall public cloud market will grow by 16.5% to be worth $203.9bn by the end of 2016.
This uptick in demand for off-premise services will put pressure on service providers' hardware infrastructure costs at a time when many of the major cloud players are embroiled in a race to the bottom in pricing terms, meaning innovation is key.
On the back of this, The Open Compute Project (OCP) is slowly (but surely) gaining traction.
Now in its fifth year, the initiative is designed to facilitate the sharing of industry know-how and best practice between hardware vendors and users so that the infrastructures they design and produce are efficient to run and equipped to cope with 21st century data demands.
Over time, a comprehensive portfolio of products have been created with the help of OCP. For the uninitiated, these offerings may appear to only suit the needs of an elite club of hyperscalers, but could they have a role to play in your average enterprise's infrastructure setup?
To answer this question, it is time to bust a few myths around OCP.
Myth 1: Datacentre efficiency is all that matters to OCP
This is largely true. After all, the mission statement of OCP founder, Facebook, was to create the most efficient datacentre infrastructure, combined with the lowest operational costs. The project encompasses everything from servers, networking and storage to datacentre design.
The server design is primarily geared around space and power savings. For example, many of the servers can be run at temperatures exceeding 40C, which is way higher than the industry norm, resulting in lower cooling costs.
This efficiency adds up to an important cost saving and a smaller carbon footprint. When Facebook published the initial OCP designs back in 2011, they were already 38% more energy-efficient to build and 24% less expensive to run than the company's previous setup.
Myth 2: Limited warranty
Most OCP original design manufacturers (ODMs) offer a three-year return to base with an upfront parts warranty as standard. This can often be better than what is offered by other OEM hardware vendors today.
The warranty options do not stop there. Given the quantities most customers purchase, vendors are open to creating bespoke support and SLAs.
In recent times, some of the more mainstream players have got in on the action. Back in April 2014, HPE announced a joint venture with Foxconn, resulting in HPE Cloudline servers aimed specifically at service providers.
Myth 3: Erratic hardware specifications
Whilst specifications do indeed evolve, the changes are not taken lightly. Any specification change is submitted to the OCP body for scrutiny and acceptance.
The reality of buying into the OCP ecosystem is that you are protecting yourself from vendor lock-in. Many manufacturers build the same interchangeable systems from the same blueprints, thus giving you a good negotiation platform.
That said, there is a splintering of design. A clear example is difference in available rack sizes.
The original 12-volt OCP racks are 21-inches but - more recently - 'OCP-inspired' servers have emerged that fit into a standard 19-inch space.
Overall, this is positive as you can integrate OCP-inspired machines into your existing racks, which has created a good transition path for datacentre operators looking to kit out their sites exclusively with OCP hardware.
Google's first submission to the community is for a 48V rack which would create a third option. But surely this is all healthy?
Google estimate this could have energy-loss savings of over 30% compared to the current 12V offering, and who would not want that? There are also enough ODMs to ensure older designs will not disappear overnight.
Myth 4: OCP is only for the hyperscalers
Jay Parikh, vice president of infrastructure at OCP founder Facebook, claims using OCP kit saved Facebook around $1.2 billion in IT infrastructure costs within its first three years of use, by doing its own designs and managing its supply chain.
Goldman Sachs have a 'significant footprint' of OCP equipment in their datacentres, and Rackspace - another founding member - heavily utilises OCP for its OnMetal product. Microsoft is also a frequent contributor and runs over 90% of their hardware as OCP.
Additionally, there are a number of telcos - including AT&T, EE, Verizon, and Deutsche Telekom - that are part of the adjacent OCP Telco Infra Project (TIP).
Granted, these are all very large companies but that quantity of scale drives the price down for everyone else. So, if you are buying a rack of hardware a month, OCP could be a viable option.
Opening up the OCP
In summary, the cloud service industry has quickly grown into a multi-billion dollar concern, with hardware margins coming under close scrutiny.
The only result can be the rise of vanity-free whitebox hardware (ie hardware with all extraneous components removed). Recent yearly Gartner figures show Asian ODMs like Quanta and Wistron growing global server market share faster than the traditional OEMs. Nevertheless, if Google is one of your customers, it is easy for these numbers to get skewed.
Even for those not at Google's scale, the commercials of whitebox servers are attractive, and it might give smaller firms that are unable to afford their own datacentre a foot in the door.
However, most importantly, the project has also led to greater innovation and that is where it really gains strength.
OCP brings together a community from a wide range of business disciplines, with a common goal to build better hardware. They are not sworn to secrecy and can work together in the open, and that really takes the brakes off innovation.