2008-09-23 Readings - Routers

Did this NOT get published on time .... ?!?!?!?

Dude saz saved not published >.< style="font-weight: bold;">SCALING INTERNET ROUTERS USING OPTICS

Isaac Keslassy, ShangTse Chuang, Kyoungsik Yu, David Miller, Mark Horowitz, Olav Solgaard, ick McKeown

Summary: Talks about the architecture needed to build a 100Tbps router. Has order 10Gbps line cards. Proposes a setup where data passes through initial queues, virtual output queues, optical switching back plane, output queues. Main contribution is an argument about how the traditional cross-bar architecture just won't scale. "Speed independent" wave grating array thing really amazing - switching speed limit is no longer RC delay for buffers/lines, but Nyquist limit for modulating a light wave!

Background: Need basic router design knowledge. At least passing familiarity with cross-bars, head-of-line blocking, etc. Also some ideas about transistor circuits and optical stuff (like WDM) would be helpful.

Discussion & Criticism: Wouldn't it be awesome if the entire backbone is optical? Like optical end-to-end. With optical slow light buffers, optical switches, optical CPU (?!?!?), and electrical-optical transducers only at the end hosts.

The paper is a bit ... dense ... Had to trust that they did their due diligence with regard to proofs etc. Their "intuitive" explanations are not all that intuitive. Do router designers need to know all the stuff re transistors and optics? Really low level!!!!

Would be nice to have a photo of these things, with the line cards, cross bars etc. identified. Would be even more awesome to have a labeled photo showing a line card with all the subcomponents, like buffers etc.

This paper should be re-read after reading the 2nd paper.


A FAST SWITCHED BACKPLANE FOR A GIGABIT ROUTER

Nick McKeown

Summary: Talks about how to build a 10Gbps router. Marks the watershed from shared-bus routers to cross-bar routers (?)

Background: Need basic router design knowledge. At least passing familiarity with cross-bars, head-of-line blocking, etc. Also some ideas about transistor circuits and optical stuff (like WDM) would be helpful.

Discussion & Criticism: Wouldn't it be awesome if the entire backbone is optical? Like optical end-to-end. With optical slow light buffers, optical switches, optical CPU (?!?!?), and electrical-optical transducers only at the end hosts.

The paper is a bit ... dense ... Had to trust that they did their due diligence with regard to proofs etc. Their "intuitive" explanations are not all that intuitive. Do router designers need to know all the stuff re transistors and optics? Really low level!!!!

Would be nice to have a photo of these things, with the line cards, cross bars etc. identified. Would be even more awesome to have a labeled photo showing a line card with all the subcomponents, like buffers etc.

This paper should be re-read after reading the 2nd paper.