As part of this Cloud 2020 series I've been looking at how the cloud will develop over the next decade. So how does the datacentre itself fit in with all this change? Is a robot-run datacentre achievable at some point in the future?
I asked Facebook's VP of hardware design and supply chain, Frank Frankovsky, where he thought the datacentre could develop next.
"I've always envisioned what could we do with a datacentre if humans never needed to go into the datacentre," Frankovsky says. "What would a datacentre look like if it wasn't classified as a working space? What if it looked more like a Costco warehouse?"
Frankovsky's ideas are informed by two things: his previous stint as head of Dell's skunkworks Datacentre Solutions Division, which sold custom equipment to massive cloud companies, and his role as chairman of the Open Compute Project — an ambitious scheme initiated by Facebook to design its own server, storage, rack and potentially networking gear, with a view to publishing these designs for the rest of the IT community to use.
Though Open Compute is in its early days, the direction its equipment is taking breaks with the approaches popularised by major enterprise vendors. Instead of designing very high-performance servers with many 'value-added' features, Facebook has gone the other way and come up with a server specification that is lightweight and maintenance and performance-per-watt above everything else.
The company also has a commitment to designing technologies that let it control many of its servers with (frequently open-source) software, breaking from the closed management tools offered by IBM, Dell, HP and others. "That gratuitous differentiation that occurs in the system management space is going to go away over time," Frankovsky says.
Over time Facebook expects that it will use more and more Open Compute-style equipment and software. Eventually, this could let it change the way it builds datacentres. One area the Open Compute Project is concerned with is "making sure the components are available at the front of the rack rather than the back of the rack", Frankovsky says.
No more people
Eventually Facebook would like to "change the way the mechanical designs work so the technicians never have to go into the hot aisle to service the machine".
And after Facebook conquers the hot aisle, it hopes its ability to manage its infrastructure mostly via software could cut the amount of time people spend on the IT floor of the datacentre — eventually, it might be possible to have no one there at all, Frankovsky says. This holds a number of intriguing possibilities for datacentres.
If people did not need to go into a datacentre, then you could deploy devices floor to ceiling and run them at a much higher heat, allowing the processors inside them to perform more efficiently, Frankovsky says.
Looking further ahead, the datacentre could be treated as a "degrade and replace" model, Frankovsky says. "Essentially, you fill up a datacentre, put it into production and weld the door shut." If a company did this, it would only need to send someone into the facility every six months to perform processor upgrade and swap out failed storage, he says.
Ultimately, if the hardware and the software gets developed, a lights-out datacentre "may be a realistic model" by 2020, he says.
A robotic future
Looking further ahead, Frankovsky imagines "a truly automated warehouse environment" where machines automatically service and swap out hardware. However Frankovsky said that this type of datacentre lies beyond 2020.
A clue to how the robotic future could develop can perhaps be seen in Amazon's acquisition of Kiva Systems in March.
Kiva specialises in robotic technologies to make warehouses more efficient. Amazon bought Kiva in an attempt to deal with some of the logistical problems brought about by the scale of the warehouses from which it ships its retail products.
The techniques applied by Kiva Systems to managing scale — automation, the use of robots to transport equipment, embedding sensors and software management into as many components as possible — deal with the same sorts of problems that a company would face in getting a robot datacentre off the ground.