Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Over time, the technology industry has developed and popularized variations of the fat tree structure. But the design has room for improvement. They’re generally reliable, but they’re also rigid, inefficient, and require complex cabling. As in actual physical cables.
If you’ve been in a data center or an office server room, you’ve probably seen nests of colorful cables spilling out of metal racks. Cabling is one of the biggest costs in networking, Rehder says, and Amazon’s global data centers are currently connected by 20 million kilometers of fiber-optic cable. This is approximately the distance it takes to travel from Earth to the Moon and back 25 times.
In 2012, with the growing demand for cloud computing services, a group of researchers at the University of Illinois Urbana-Champaign, including Godfrey, presented It is a concept known as jellyfish. The fixed network designs in use at the time were struggling to meet growing demand, so the researchers proposed a “high-capacity interconnection network which, by adopting a random graph topology, naturally leads to gradual scaling.” They believed that this random approach could be more efficient and scalable than networks built using a fat-tree structure.
“We called it jellyfish because it’s liquid,” Godfrey says. “You can connect routers and switches randomly and it becomes this elastic pooling of network capacity, and it’s very efficient.”
However, Jellyfish also presented new challenges in layout, data routing and cabling. Routing in random graphs is more difficult, Godfrey says, because there are many diverse paths the data can take from its source to its destination. Connecting cables is more difficult because the endpoints of the cables are chosen randomly.
Two years later, Google started thinking about another solution: this Began to integrate switching optical circuitsor OCS, in its network designs. This approach uses small mirrors to reflect light from the input port to the output port, allowing Google to reconfigure the optical cables in real time. But, again: this adds a certain amount of engineering complexity, as well as cost.
Courtesy of Amazon
.gif)
Courtesy of Amazon
Very random
Meanwhile, Amazon was searching for the “Holy Grail,” says Giacomo Bernardi, one of the lead authors on the new paper, along with Amazon researchers Ratul Mahajan and C.S. Sechandri. In an ideal world, the data network would be flat and efficient, resilient to hardware failures, random enough to maximize performance, and scalable enough to grow without becoming unwieldy. It will also rely on simpler, more streamlined cables rather than increasingly complex fiber-optic systems.
When he and his colleagues began trying to build such a network, Bernardi says he really became obsessed with Penrose tiling, a type of aperiodic tiling named after the British physicist Roger Penrose. (Other researchers It was very inspired by the Penrose court (They tried to translate the patterns into error-correcting code in quantum computers.) Bernardi wondered if Amazon could use a similar construction and create a flat “lattice” following a repeating pattern. He and his team tried to build a simulation of what that might look like.
