why an has always been doomed laurent mathy lancaster university

7
Why AN has always been doomed Laurent Mathy Lancaster University

Upload: brett-turner

Post on 14-Jan-2016

213 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Why AN has always been doomed Laurent Mathy Lancaster University

Why AN has always been doomed

Laurent Mathy

Lancaster University

Page 2: Why AN has always been doomed Laurent Mathy Lancaster University

What is AN?

• Remember it all started with capsules

• To many this has evolved towardsstore-process-forward

– User-controlled operation

• This all means per-flow processing– Complex packet classification– Micro-flows or small aggregates– Memory/cpu bottleneck

Page 3: Why AN has always been doomed Laurent Mathy Lancaster University

AN: where?

• As a result, AN where flows are “scarce”

• That means the outer edge of the network– Either inside tier-3 ISP at access with

customers– Or inside customer networks

(campus/corporate networks)– A no-no at tier-3/tier-2 boundaries and deeper

• This is near end-to-end

Page 4: Why AN has always been doomed Laurent Mathy Lancaster University

AN: usefulness

• Because AN is only viable at outer edges, usefulness is rather limited– Why push complex processing at

• End of first hop?• Start of last hop?

– Can still be useful to help cross-layer designs• Allow applications to “peek” inside the network

– Measurements (speed, volume!)– routing info, etc– Video transcoding anyone????

Page 5: Why AN has always been doomed Laurent Mathy Lancaster University

This guy is wrong!

• “Wait! AN is open node architecture, open interfaces and clever software management (modular systems, reflections, etc)”

• But OpenSIG/OpenArch have failed– Do you remember IEEE PIN 1520 ?

• Node architects are doing fine, thank you• Software engineers, OS developers and

the others haven’t been sleeping– modprob!

Page 6: Why AN has always been doomed Laurent Mathy Lancaster University

Imagine…

• Even if CISCO can be persuaded and Network admin let’s you run “vetted” third party code… and something’s not working– Who you gonna call?– You have compatibility issues with the other

“side”• Looks like there are too many people who can

blame each other

– After too many disturbed lunches:active-net: disable

Page 7: Why AN has always been doomed Laurent Mathy Lancaster University

Conclusion• Wherever I look, I see issues for AN

– Not much operational sense• Core today is edge tomorrow

– Not much commercial sense• Constructor’s protectionism

– Academically, contribution has been to mostly show what not to do

• It is doomed and always has been…please leave it where it belongs