Amazon Thinks Future of Data Centers Depends on Technology Problem Just Solved

Over time, technology companies have developed and implemented variations on the oil price architecture. But the design has room for improvement. It is often unreliable, but also fixed, inefficient, and requires complex cabling. Like, real, real strings.

If you’ve ever been in a data center or the server room of an office building, you’ve probably seen nests of colorful cables running out of racks. Cabling is one of the most cost-effective aspects of networking, Rehder says, and Amazon’s global data centers are currently connected by 20 million kilometers of fiber optic cables. That’s the distance it would take from Earth to the moon and back 25 times.

In 2012, when the demand for computer services is growing, a group of researchers from the University of Illinois Urbana-Champaign, including Godfrey, launched. a concept called Jellyfish. The fixed formats that were used at the time were struggling to meet the growing demands, so the researchers decided that “high-level connections that, using graph topology, naturally lend themselves to increase.” They believe that this random method can be more efficient and riskier than a network built using oil prices.

“We gave it the name Jellyfish because it’s fluid,” says Godfrey. “You can connect routers and change them randomly and it becomes a dynamic network pool, which is very useful.”

However, Jellyfish also introduced new challenges in layout, data paths, and cabling. Running random graphs is easy, Godfrey says, because there are many and varied paths the data can take from its destination. Cabling is very complicated because the ends of the cables are randomly selected.

A few years later, Google started toying with another solution: It began to combine the light circuit to changeor OCS, in its network design. The system uses tiny mirrors to reflect light from the input to the output, enabling Google to visualize cabling in real time. But, again: This adds technical complexity, and cost.

Courtesy of Amazon

Courtesy of Amazon

That’s Random

Amazon, meanwhile, was looking for “cleanliness,” says Giacomo Bernardi, who is one of the authors of the new paper, along with Amazon Scholars Ratul Mahajan and CS Seshandhri. In an ideal world, data networks would be flat and efficient, tolerant of hardware failure, random to improve performance, and scalable to scale without degradation. It would also rely on simple, streamlined systems rather than complex, sophisticated systems.

When he and his colleagues started trying to create such a network, Bernardi says he was already fascinated by Penrose tiling, a type of aperiodic tiling named after the British scientist Roger Penrose. (Other researchers was strongly inspired by Penrose tiles that they will try to translate the patterns to correct errors in quantum computers.) Bernardi wondered if Amazon could use a similar architecture and create a flat “mesh” following an iterative process. He and his team tried to create a simulation of what it would look like.