Wednesday, May 26, 2010

bell heads v. Internet: energy systems, transport and communications

So we kind of won the war between Bell-Heads of the phone companies and the Net-Heads of the Internet in the communications area - indeed, now we made peace and we are very happy to work with each other on better inter-domain routing, delay bounds, congestion exposure, traffic engineering etc etc etc

Meanwhile, Frank Kelly spent time at the UK's Department for Transport as chief scientific advisor. More recently, David Mackay is acting as same role for the Department of Energy and Climate Change

Why is this interesting (in the context of the Net)?

Well, this week, I am attending the Energy Systems week at the Newton Institute in Cambridge and the talks have fallen in to the same pattern of three sides of the picture that we saw in the net, and transport: Energy is dominated by a shall number of energy producers and grid companies (mostly private nowadays). THe move to local generation
(i.e. "user contributed" energy) and the move to using prices as "signals" to users to see if that will alter their (consumer and producer) behaviour have been things we've seen in the net (p2p content versis Content Distribution services, congestion charging and congestion explosure as well as Inter-domain policy routing/peering policies), and in Transport (congestion charging, toll v public good roads).

Glen Vinnicombe gave a great talk about why distributed control with local rules for interaction is sufficient to organise a globally stable system. Alas, it was clear from many responses that the industry (i.e. people that have outrageously expensive systems invested in command & control - 100M pounds software for running LP and IP solutions for optimisation was one quoted figure!!!), have no understanding of why we might need to move to this decentralised world.

What we really need is a subtle political move that shifts the ground under their feet, as was done in the Transport case (allowing local decisions to innovate for example so that people could introduce smart pricing, smart control independent of the business dominated by near monopolies). This move needs to happen soon for energy - it is not at all clear how to lay out a vision and road-map for this that could engage the incumbents - in the case of the internet, we just ignored them (as Americans say, we did an "end run" on their game:).

Putting in constraints (simple things like Kirchhoff's laws is trivial compared to the complexity that is the Internet. I was amused that people think the electricity system is harder than a global packet switched system with over 100,000 providers. (Other things like the fact that we have different kinds of packets (VOIP/Video v. Data), the fact that intra-domain TE runs on timescales faster than spot markets (as do CDNs, who buy electricity dynamically and spin up/down disk farms and whole city datacenters already based on current supply cost/reliability, and buy wind farms) seems to have slipped the attention of some of the old school....

Anyhow, thanks are due to the Newton Institute event organizers for bringing together a set of people to make this clear.

Some brief other notes-

  • Some of the economics people were still citing Chicago school work on pricing and perfect markets - having heard of Cascading failures in grids, you'd think they'd have heard of Cascades and herding in trading behaviour and move on a bit to worry about whether you can actually treat users (and providers) as selfish rational players and rely on this for stability alone - the last 2 years recession (toxic debt and trust failure epidemics) ought to have been another small clue.
  • Thatcher mentioned as a contribution - its true the pool idea was cool, but society was a loser compared to some more balanced (nuanced) inclusion of welfare as well.
  • Glen's demo of the lego robot segway was absolutely brilliant:-)
  • Dinosaurs are not extinct
  • Andy Hopper is right - Computing for the Future of the Planet actually has something serious to offer here.
  • There was some interesting discussion later in the week on large scale microgeneration and stability, but there still seems to be a misunderstanding of the sorts of scale (millions of providers) and the types of business relationships between them (P2P, customer provider, laissez-fair etc) - BGP has many lessons here potentialyl of use
  • there was some discussion of the unpredictability of weather (for wind & solar) - in fact systems are self-similar and the same math used for rapid estimation of the H parameter for a self-similar arrival process (lots of papers on fractional guassian random/brown noise etc etc) could easily be used to provide input to an "admission process" of local generation to the grid, or whetehr to diffuse the generated power locally (e.g. into bufffers in the form of electric cars parked in a neighbourhood, or to spin up disk farms for a decentralised local cache for the InterWeb).

    I've always said that the Internet is a predictor for where other utilities might end up (given its scope, scale, low operating costs, competitive nature, and flexibilty to a wide varieity of ever evolving business models)
  • Sunday, May 23, 2010

    simple smart grid idea - turning down power consumption via missing NAT state:)

    If you're home is not currently represented by NAT state, then surely all your internet kit should be OFF :-)

    Thursday, May 20, 2010

    An ethical question about live user internet research

    we're running a complicated experiment to see how people react to information about
    their social group and their locale during an epidemic if you give them live information about who (and potentially where) there is some level of infection - this had to go through an ethics committee because it involves health.

    now the next phase (or a next phase) involves running an enhanced version of the programme which no longer logs whether a person has a disease, but instead, emulates a set of diseases and a set of values for parameters for Susceptibility, Infectiousness and Recovery, and (still telling the user - but now its a game) tries to see if people will alter their behaviour (i.e. where they go; who they meet) depending on the severity etc - the idea of the latter experiment is that we can run live experiments with hypothetical values for the disease vector and SIR values - the reason we can't just run this on trace data is that there are feedback loops between the disease and the social network dynamics, so whether a given virtual epidemic collapses, stays endemic, or goes pandemic (or some other mode) will be highly dependent on actual users' behavioural dynamics (and might include other facets of behaviour like peer pressure etc)...

    Anyhow, the question is this: do we still need formal Ethical approval for this latter experiment?

    Friday, May 14, 2010

    Internet Addressing - making a hash of it several times

    Dear Sir, Lord, Reverend, Almighty Internet,

    I've been wondering how you should be addressed.

    The powers that be (wise old men) say we should have 128 bits of stuff
    allocated either geographically or topologically, or organisationally (provider centric)

    I say we should declare IPv6 Bloomsday and have three hash functions and addresses should be allocated with fields set from all three types, by doin a cryptohash of the ID of a node (sim or mac) ...

    So the new type of address is the bitwise inclusive OR of the results of all three hash functions:
    Hg(id) v Ht(id) v Ho(id)

    The properties of this are neat - a router can decide to apply a match on any one of the three hashes, so if a node moves, we can see whether geo, topo, or provider based routing will cope with the migration best - we can also NAT or re-assign permanentally for a node, any 2 of the hashes, and still get a match - provided we dont change all three (or the id)...

    This ought to work fairly well, since most nodes execute Levy flights over the net (returning to the nest at end of day, trip, vacation)....

    Saturday, May 01, 2010

    Blog Archive

    About Me

    My photo
    misery me, there is a floccipaucinihilipilification (*) of chronsynclastic infundibuli in these parts and I must therefore refer you to frank zappa instead, and go home