Wednesday, February 4, 2015

IP Multicast : the Revival ?

IP Multicast: a revival in smart grids?

Since its inception in the early 90's, IP multicast has been confined to local area networks and to TV distribution systems. With IP multicast, it is the routers that duplicate messages, not the sources or intermediate servers. This makes the job of the source much easier: with IP multicast, an IPTV server sends only one stream, even if there are 1'000'000 receivers. The magic is in that if there is in average 10 hops from source to destination, and if every router duplicates every packet of the stream 4 times (a modest piece of work for a router), then at the end there are 4 to the power 10 (ca. 1'000'000) duplicates reaching the destinations -- in reality this is never as simple as multicast trees are not balanced, but this gives the idea.

So if multicast is so powerful, why is it so little used ? Perhaps for several reasons. First, multicast needs a multicast routing protocol and there is overabundance of them, making it hard for an operator to choose the right one. Perhaps this problem will go away with software defined networking (SDN), which could replace multicast routing protocols in a much easier way. Second, multicast has a bad reputation when it comes to security: denial of service attacks are much more powerful if an attacker finds a way to use multicast (here, too, SDN will help). Also, widespread end-to-end security solutions such as DTLS do not support multicast.

Multicast IP is a natural solution for smart grids, where many simple sensors (such as Phasor Measurement Units, PMUs) stream measurement data every few milliseconds to multiple destinations. With multicast, you configure one single destination address in a PMU and then forget about it; if a new destination comes and listens to the data, no change in the PMU is needed. In contrast, without multicast, the PMU would need to start a new session for every additional destination.  In today's electrical substations, multicast is used at the MAC layer. In tomorrow's smart grids (which will be wide-area networks), this will not be sufficient and IP multicast will be needed. A good opportunity to solve the security problems posed by IP multicast ? Carl Gunter and Jianqing Zhang at University of Illinois have tackled the problem, see Jianqing's dissertation.