The FireFly architecture combines the benefits of having a low transmission power and a small interference footprint, according to researchers.
Using wireless optical links steered via MEMS-mirrors, a team of computer scientists has devised a networking architecture for data centres that is dynamically reconfigurable to accommodate inter-rack traffic in the best possible way.
Their paper, "Novel architecture for reconfigurable optical wireless networking data centres," published in the Society of Photo-Optical Instrumentation Engineers (SPIE) Newsroom, highlights the need for flexible and adaptive network fabric in data centres while offering a lightweight and efficient solution without overprovisioning wired links.
Circumventing complex static cabling schemes, the researchers envision a network architecture that uses free-space optics communication links to create an all-wireless inter-rack fabric capable of supporting data rates in the tens of gigabit/s across large computer farms (with distances over 100m). According to the authors, the novel architecture dubbed FireFly would combine the benefits of having a low transmission power and a small interference footprint.
In the FireFly architecture, each top-of-rack (ToR) switch would have steerable free-space optical links able to connect to other ToRs, flexibly, to adapt the network to changing traffic workloads.
__Figure 1:__ *Schematic illustration of the FireFly architecture. FSO: Free-space optics. ToR: Top of rack.*
To prove the feasibility of their concept, the researchers have built a proof-of-concept prototype consisting of MEMS mirrors (capable of a 10º optical deflection) steering a collimated laser beam through a wide-angle lens (magnifying the optical scan angle to over 30°). This steering mechanism draws less than 1mW and operates at up to 1.2kHz.
__Figure 2:__ *The MEMS-based proof-of-concept prototype assembly used to realise steerable FSO beams.*
The researchers also developed practical heuristics to address the algorithmic and system-level challenges in the network design and management of such architecture. They acknowledge that they'll need specific algorithmic techniques to make these flexible networks a reality, including algorithms for the joint optimisation problem of runtime topology selection and traffic engineering, as well as data-plane mechanisms to guarantee various consistency and performance requirements.
They are now busy building a small testbed for the FireFly architecture, including auto-alignment through the use of galvanometers and MEMS steering technologies.
This article first appeared on EE Times Europe.