“Everything is getting better with Moore’s Law, but some things are getting better faster than others.”
— Kenneth Church, “The Mobility Gap”
Almost everyone has heard of Moore’s law, which states that CPU processing power goes up by about 100x per decade. Fewer of us have heard of Kryder’s Law, ie the “Moore’s Law of storage” – it says that we can expect a 1000x increase in disk capacity per decade. And even fewer have heard of Nielsen’s Law for networking – which only claims a 256x increase of bandwidth capacity in the workplace over the same period.
In short: bandwidth rates aren’t keeping up – in a large part due to limitations of the speed of light. It is this gap in improvement in storage and bandwidth density that Mr Church coined the “mobility gap”. How big is the gap? Roughly 4X – 18x over each decade. Which doesn’t seem that bad until you also consider the “digital universe” problem which says that AMOUNT of global data is 44x time what is was in 2009 by 2020.
Researchers are particularly hard-hit by this growing problem. As instruments get more and more sophisticated, the amount of raw data that is capture is increasing all the time. At the same time, more and more scientists around the world want to get their hands on it for their own research, meaning that huge amounts of raw data are being shipped around the world.
So we have MORE data, LESS ability to move it around, and MORE demand to analyze it.
One of the best ways to increase bandwidth levels is to improve the utilization rates of existing circuits. And that is exactly what researchers at Caltech and CERN did. At this years’ Supercomputing Conference (SC12) exhibition in Salt Lake City from November 10-16, they demonstrated by using a per-flow multipath switching fabric based on Floodlight.
The team, lead by Dr Michael Bredel of Caltech and CERN, implemented various Floodlight modules that form a basic per-flow multipath application, and extended the given topology management to calculate link-disjunct paths between source and destination nodes. The corresponding OpenFlow forwarding entries are pushed to the related switches automatically, whenever a new flow appears at the ingress of the OpenFlow network. Moreover, new flows are allocated to possible paths in a round-robin manner.
To demonstrate this, they created a world wide meshed layer-2 OpenFlow overlay network by connecting Pronto OpenFlow switches in Geneva, Amsterdam, Chicago, and Salt Lake City via 10 Gbps links on top of a standard R&E network environment. The direct connectivity of the OpenFlow equipment has been achieved by VLANs, and they used multipath TCP (MPTCP) to split a TCP datastream to multiple TCP sub-streams and send the data on multiple paths from source to destination.
They proved it is possible to improve the datarate, and push it beyond the limits of a single path. Another great example of SDN improving ROI.