Intel Labs Builds Routers From Clustered Servers

Intel Labs claims that it has come up with a way to reuse and/or reconfigure commodity-type servers and cluster them in such a way as to turn them into data centre routers.

Router bricks

Intel researchers, led by Gianluca Iannaccone and Sylvia Ratnasamy, have coined the phrase “router bricks” for the reused servers, which are designed to put servers that may be out of commission – or new units not being used right away – to work in new capacities, thus saving capital expenses for IT departments.

Key to the redeployment of these machines is an open source software package called Click Router, developed at MIT a decade ago, which the Intel researchers have used to tie the servers together for their new roles in the data centre.

Router bricks are a high-speed router using off-the-shelf IA [Intel architecture] servers. They are fully programmable [control and data plane], extensible in that they evolve networks via software upgrade, and incrementally scalable at flat cost per bit, Iannaccone told eWEEK.

“These are a first step toward flexible network infrastructure,” Iannaccone said. “We are currently pursuing the application of RB to data centres, where they will help in content delivery, network power management and next-gen Internet routing.”

Nehalem chips enable enterprise-level performance

These create networks that are simpler to use and cheaper to evolve, Iannaccone told eWEEK.

“Programmers can rapidly build and reprogramme networks using the hardware and software they’re most familiar with. They [also] can decouple network software and hardware and avoid the cost of specialised hardware development,” Iannaccone said.

The main reason this can be done at this time is the emergence of Intel’s multi-core Nehalem chips, which provide the bandwidth and gigbit speed for these router bricks to perform at enterprise levels, Iannaccone said.

“The router bricks demonstrate that any number of servers can achieve switching speeds of N × R bits-per-second, provided each server can process packets at a rate between 2R-3R bps,” Iannaccone said.

“The Nehalem chips have the power to do this. We couldn’t have done this before.”