Tuesday, July 10, 2018

Variational Principles

Amazon link

Like Ted Chiang ('Arrival'/'Story of Your Life') I am a huge fan of variational principles. I was therefore pleased to discover Jennifer Coopersmith's well-regarded book (above). She has a good article about the connection between Newtonian Mechanics and the Principle of Least Action which I stumbled across when I was thinking about something quite different .. .

---

First Brexit. Beneath the daily froth of politics find the plate tectonics of political-economy. Political forms and institutions seek their stable equilibria, their maximal-entropy configuration. There's a variational principle here!

The UK is a small island abutting a Franco-German empire . The EU likes to drape itself in high-falutin' ideals but it's really about elite power and economic efficiency. Institutionally it's in crisis. Neoliberal stagnation and decay has frayed the bonds of social cohesion across the western world. Yet the economics of international supply chains, specialisation and scale permit no national regression. The consequent collapse-process is slow and uneven - and BRINO was perhaps always the fated outcome in this phase.

---

I was also thinking of my days as a BNR network planner in the 1990s, bringing modernity to the ex-socialist states of Eastern Europe. I was a frequent flyer, regularly presenting Nortel's model for modern network design to Polish, Hungarian and Romanian telecom planners.

We would take a country and look up the main town and cities in an atlas. We'd carefully measure the distance between every pair of towns to create a distance matrix. Then we'd check the populations figures: so many thousands of people per city, and apply a multiplier to work out the total traffic generated (erlangs in those days, now G/Tbps).

Next came the clever part. We used a gravity model on the inter-city distance matrix to work out how much traffic each link would have to carry. As I recall, we didn't use an inverse-square law: empirically that was too sharp a cut-off. A 1/distance rule worked better. So now we had a traffic matrix.

This was input into our network design tool, and out came design-graphics and equipment lists for access, aggregation and long distance core networks. Our audience, people who'd been using pencil and paper hitherto, were awestruck.

We made a lot of sales.

I mention this only because our solutions, too, seemed instances of a variational principle. The optimal cost-benefit solution for a network design, given the 'traffic potential field' defined by underlying populations and geographic-separation parameters.

---

Update: Gwern has researched all this to within one inch of its life.


6 comments:

  1. I assume that these gravity models were for telephone traffic rather than internet traffic? Nor was this really gravity...

    (Roy - sent to test Blog Comments system.)

    ReplyDelete
    Replies
    1. Back in the nineties .. yes, it was telephone traffic. How we worried about the appropriate statistical (Poisson?) distribution for data!

      Delete
  2. Good that the comments are working again (although I cannot comment on the latest post itself). Nevertheless this is an opportunity to question whether Nortel's failure was simply due to "non-Culture managers" arriving - I always thought it was primarily due to Nortel's non-adaptation to the Internet era.
    The fact that even in the late nineties Nortel was only considering voice traffic (by which time Google, Yahoo, Amazon were getting founded) is a hint.
    The non-Culture managers probably did arrive as (US) Internet companies were hastily bought in resulting in a mix that never settled.

    In WikiP the 2000s are known as the Nortel era of "The Right Angle Turn" - I wonder what that refers to?

    Incidentally a recent BBC4 program on the rise and fall of Nokia, had Nokia staff also blaming their fall as due to non-Culture managers coming in at the height of their success.

    ReplyDelete
    Replies
    1. I remember BNR Div 6 (systems engineering) meetings in the late nineties strategizing the Internet. Cisco was seen as owning enterprise IP, especially routers while Nortel was a trusted supplier to public network operators (super high reliability) in voice.

      It was seen as a race to the future middle. The meetings were intense. Nortel bought a small Cisco competitor, Bay Networks. And within a decade Cisco had won decisively.

      Delete
    2. Interesting. But I wonder whether Cisco were acting alone ie the challenge for Nortel systems engineers was to "integrate" its voice networks (ultimately 7-layer OSI based, I believe) with the TCP/IP world. No wonder the meetings on this were intense (and as we now know unsuccessful).

      Cisco were approaching all this from the TCP/IP side and would have had "friends" they could link to. There could be a lot of Intellectual Property and Patents on this issue (OSI to TCP/IP) that neither Nortel nor Bay Networks possessed, but they needed to develop quickly. I suspect that it never happened.

      Delete
    3. Nortel was *amazed* to discover that carriers *queued up* to deploy high-end Cisco enterprise routers (7500s) in their urgency to offer Internet services. Nortel never got off the back foot, and its increasingly frenzied efforts to transform itself (that right-angled turn) culminated in hiring those toxic new C-level managers.

      Followed by Chapter 11.

      Delete

Comments are moderated. Keep it polite and no gratuitous links to your business website - we're not a billboard here.