When I imagine “the cloud” where most of my daily interactions with web services such as Facebook and Google compute, I think of a free-floating space — complete with scalloped edges — that isn’t tied to a physical location.
In reality, however, cloud computing is very much tethered to the physical world, through football-field-sized warehouses of connected servers that enable it. A computing job, though it can be accessed from anywhere in the world, can’t be easily moved from one warehouse to another. It can’t even necessarily be moved from one area of a warehouse to another area of the same warehouse.
This is a problem for two reasons. The first is that it’s inefficient and drives the price of cloud-based applications that use data centers upward. When a new processing job comes up, it can’t just jump to any idle server in the warehouse. Rather, it needs to use a specific location in the network of servers.
Martin Casado, whose PhD dissertation at Stanford University addressed this issue, estimates that for this reason most data centers run at about 40% to 60% of their capacities — costing companies tens of millions of dollars and slowing down their services.
The second problem was most recently demonstrated in the wake of the massive earthquake and tsunami that devastated northern Japan last year. After the disaster, most of the region was put on a schedule of rolling blackouts to cope with the energy shortage that accompanied the tremendous loss of life and property.
Many of Japan’s manufacturers shifted production outside of the Tokyo area to deal with the energy shortages. Its data centers, however, could not, and many relied on diesel generators to stay in business.
“You have these huge warehouses of computers doing all of these important workloads,” Casado says. “If you actually do a power outage there, you’ll lose all of them. If the cloud was really as elastic as we say it is, we should be able to take all those workloads and move them to another data center.”
But why can’t we? The short answer is that workloads are still attached to the network of computers they’re being processed on.
Why is this such a hard problem to solve? Networks of computers get most of their functionality from hardware, which is hard to update in a way that would allow the flexibility needed to move workloads from one data center to another.
“Whenever you bake something into hardware, it takes four years to design, you spend millions of dollars and you can’t really change it,” Casado says.
In order to untie the cloud from a physical location, Casado thought up a new approach that hadn’t really been tried before: Software.
The startup he founded, Nicira, based on his doctoral dissertation makes a program that moves network functionality into the software rather than the hardware, which gives it the flexibility to move workloads around. He compares it to the film The Matrix.
“When you have an application or a workload, and it’s running in the cloud, what we do is move that application in to The Matrix,” he says. “So it doesn’t know it’s not running in the real world. It’s running in this totally generated world — we can move the entire matrix from the United States to Japan, and the workload has no idea the world has changed.”
We’re satisfied with the movie metaphor. If you’re interested in the nitty-gritty details of how this works, check out Nicira’s explainer video.
The overarching point — why Nicira was able to raise $50 million and why companies such as AT&T, eBay and Rackspace have purchased its software — is that adding this flexibility to data center workloads makes it less expensive for cloud-based services to operate. It also protects workloads from being affected by events at the data center locations and makes cloud applications run more efficiently.
Casado says that adding a new application offering to one of his customer’s services used to take seven days. With Nicira, adding the same type of applications now takes 30 seconds.
By truly freeing the cloud from the ground for the first time, Nicira could help it rise to new heights of functionality.
Sarah Kessler -Mashable