Networking equipment vs. Moore’s Law

Last week, I was at the NANOG conference in Vancouver.

The opening day’s agenda featured a thought provoking keynote talk from Silicon Valley entrepreneur and Sun Microsystems co-founder Andy Bechtolsheim, now co-founder and Chief Development Officer of Arista Networks, entitled “Moore’s Law and Networking“.

The basic gist of the talk is that while Moore’s Law continues to hold true for the general purpose computing chips, it has not applied for some time to development of networking technology.

Here. you can see that Moore’s Law is very much holding true for the general purpose CPUs which drive things such as our own computers, and for the machines which make up the massive server farms used by the likes of Google, Facebook, Amazon, et al.

The bad news is that networking technology has not really been keeping up. Look how long it’s taken us to get from 1G to 10G to 40G to 100G, and all the while, the high-speed networking user community – Clustering, Cloud Computing, R&E Networks, Internet Exchanges, the largest carriers and ‘ultra giants’ such as Google, Amazon, et al – have been pushing, adopting fast networking tech the moment it’s released.

Bechtolsheim seemed to indicate the only way that networking can pull back is by making custom “system on a chip” type forwarding engines to increase system density, providing enough ports of sufficient speed. This is a change in tack compared to 5 years ago, when a number of networking suppliers had closed down their chip design departments, instead choosing to integrate off the shelf merchant silicon into their platforms.

So, what’s been going on? Have the networking equipment companies been occupying a “reality vacuum”, oblivious to the networking needs of the devices attached to them? Lagging behind while maximising yields for their shareholders from ancient networking kit that has less memory and less processing power than the smartphone in your pocket? Sometimes it feels that way.