Researchers craft algorithm to turn mesh networking green
The energy costs of IT equipment have become a major concern, leading to a variety of creative ideas about how to keep power use in check. Most of the challenge has been finding ways to make sure that the equipment that's in use gets fully utilized, while anything that isn't gets put into a low-energy state. A paper in the Journal Of Lightwave Technology suggests that, under the right circumstances, the same thing can be applied to wireless networks, producing significant energy savings.
The focus of the research is on "last-mile" connections, which take network traffic from the local phone company's exchange to businesses and residences. After dismissing fiber-to-the-home as "cost-prohibitive," they suggest that a wireless mesh is the best way to solve many of the last-mile problems. In an appropriately dense environment, a few fiber-fed access points, either 4G or WiFi, could allow the signals to be propagated out through the networking hardware of individual users.
Mesh networks currently require a routing algorithm to route traffic from a given user through the intervening hops on other users' hardware and onto the phone company's equipment. Currently, these routing algorithms tend to focus on two things: minimizing the number of hops, and load balancing, which distributes traffic as evenly as possible. This helps ensure that no single piece of hardware gets swamped, but as the authors note, it has a downside: it also ensures that a lot of the hardware is active (and burning electricity) even when the network's overall capacity isn't being strained.
To try to do better, they designed a mesh network and routing algorithm with energy efficiency in mind. The hardware configuration is what they call WOBAN, for wireless-optical broadband access network. The optical portion isn't wireless; instead, the design would call for ISPs to install major access points at the end of some optical fiber; the authors favor passive optical networking (PON) for its low energy use. These would feed optical network units, which would offer the network via some form of wireless technology. The authors suggest both WiMAX and WiFi would work, but it would seem to be applicable to anything that can form a mesh network.
From the optical network units out, access is provided by a wireless mesh. Both the optical network units and the mesh hardware, in order to provide reliability as individual transmitters join and drop off the mesh, will have to be redundant—a given user's access point will have multiple ways of sending packets back to the ISP's fiber. Normally, with a load-balancing algorithm, a lot of this hardware would end up constantly active, since packets would be distributed evenly across all reasonable routes.
From an energy efficiency standpoint, this is exactly what you don't want. So, the authors experimented with developing an algorithm that would fill fewer routes at a higher capacity. This would allow some of the hardware, both mesh components and optical network units, to be put to sleep for extended periods of time.
The algorithm takes a variety of factors into consideration. It includes what the authors term "watermarks" of high and low network utilization. The high version ensures that the mesh doesn't try to stuff any more bytes through an over-capacity mesh component; the low provides an indication of when a given piece of hardware is a candidate to be put to sleep. The algorithm also tries to keep the number of hops between users and the Internet constant, and avoid wireless interference.
Challenges remain
Unfortunately, with all these different considerations, solving this ended up being NP-hard, which means the time involved in calculating it gets much larger as the number of nodes increases. In fact the authors attempted to model a moderate-sized network and gave up. They suggest that some approximations might help improve performance, but don't actually demonstrate any of these. Another thing that's missing is a consideration of the energy needed to put a processor to work trying to crunch these numbers; it's possible that at least some of the Watts saved would be lost to this.
In any case, the authors modeled a WOBAN in Davis, California that served a combination of university, business, and residential users. Using a watermark where utilization under 10 percent targets hardware for shutdown, the authors found that, on average, nearly two-thirds of the optical network units could be put into sleep mode using their algorithm. Power consumption dropped by dozens of Watts at all hours of the day, even as the network usage shifted from the university and business area to residential use in the evenings.
All measures of performance, such as number of hops and throughput, remained constant.
It's a bit difficult to get overly excited about these specific results. There's no clear indication that mesh networking will end up being the method of choice for providing last mile services, and the algorithm itself clearly needs a bit of work. What is interesting is the general approach; load balancing seems to have become a standard method of handing out distributed tasks, but it clearly doesn't work as well from an efficiency standpoint. It's possible that other compute tasks that are more intensive users of electricity could benefit from a similar analysis.
Journal Of Lightwave Technology, 2010. DOI: 10.1109/JLT.2010.2044369 (About DOIs).
No comments:
Post a Comment