New Internet architectures to reduce carbon emissions

[This is another posting as part of my own evolving thought processes how the Internet, and in particular research and education networks can help reduce carbon dioxide emissions, firstly by re-engineering the network and secondly by deploying applications and services that will encourage others to use the Internet in novel ways in order to minimize their own carbon footprint.

First of all I would like to thank all those people who sent me e-mails with additional suggestions, comments and ideas on how ICT technologies, in particular the Internet and broadband can be used to mitigate the impact of global warming. Given the large number of e-mails I have received on the subject I apologize if I have not been able to reply to some of you directly.

I want to assure you that none of my ideas, and those of others that have been posted here, are in any way cast in stone or anywhere close to deployment. Many of these ideas are come from my own fevered brain, and may likely never survive close scrutiny by experts or validation in the marketplace. The purpose of this e-mail and my blog is to hopefully stimulate some creative thinking in the Internet community and especially within R&E networks on ways we can collectively design "green" Internet solutions. This is a community that is used to rapid changes and has many of the most innovative people in business or academia. Hopefully my blog, in some small way, will stimulate others in developing more robust and scalable solutions that help address, what in my opinion, is the biggest challenge of this generation and of this decade - global warming.


In today's modern Internet networks one of the biggest energy sinks, and consequently a significant producer of carbon emission due to their electrical and cooling requirement, is the Internet core routers.

Internet routers are custom designed pieces of computing equipment which must operate at very high speeds in order to do fast lookups in the forwarding table in order to process packets at line speeds. The need to do fast lookups is further compounded by the continued growth of routing tables over the past few years.

In order to handle the processing of packets at wire line speeds modern routers usually have multiple ASICs on the forwarding card. Each ASIC handles only a subset of the forwarding address table, which is split up between the various ASICs on /8, /16 (or finer grained) address boundaries.

But an alternate routing architecture approach to big core routers with multiple ASICs is to deploy networks of multiple virtual routers, with each network of virtual routers assigned an address block. All virtual routers for a given address block linked together by a dedicated lightpath network independent of parallel virtual routers and networks for other address blocks.

Each address range or block would have a global set of virtual routers dedicated to forwarding and routing with that address block. And optical connections between the virtual routers can be traffic engineered to optimize flows for that address block. As well separate OSPF (or ISIS) networks can be deployed for each address block. At inter-domain boundaries these separate address block networks can be aggregated into a single connection to a neighbouring AS, or arrangements can be made to advertise separate BGP networks with parallel ASs for each address block network.

At first blush this seems to be an incredible waste of resources. Not only would separate routing tables and networks would have to be maintained, but multiple copies of filtering policies etc would have to be deployed for each network address block.

However by breaking up the forwarding table into multiple (roughly) parallel forwarding networks, where each network is assigned a specific address block allows us to deploy much more inexpensive commoditized routers using off the shelf open source routing engines like Vyatta.

Because these routers don’t have to do lookups on the entire forwarding table they can be built with more inexpensive commodity components. In effect we are trading off large forwarding tables using ASICs against commodity virtual routers with multiple parallel optical networks for each address block.

More importantly these low cost (and low energy, hence low carbon emission) devices can now be collocated nearby renewable energy sites. Not all such sites need to have to support all virtual routers to carry the entire routing table. Instead address block networks can be engineered with different topologies linking together independent renewable energy sites supporting alternate nodes for the various address block networks.

Because we have also broken down the Internet into many (roughly) parallel networks aligned along each address block, outages and re-routing can be more easily handled, especially as the routing nodes are located at renewable energy sites such as windmills and solar power farms.

Users would be backhauled to with dedicated optical links to two or more virtual router renewable energy sites. The assumption is that an all optical backhaul network has much lower carbon emissions than an energy consuming electronic local router or stat-mux switch.

This architecture would be ideal for R&E networks as generally they have a very small number of directly connected organizations such as universities and research centers. These organizations can even pre-classify their outgoing packets along the address block boundaries and send them out separate parallel optical channels to the nearest renewable energy site(s) supporting the multiple virtual routers for each address block.

Companies like Google are also well positioned to take advantage of this architecture as they have a world wide distributed network of low cost servers and they are rumored to be deploying costumed developed 10Gbe switches on their own private optical network. The same principles that Google used for their network of search engines could be applied to a virtual routed network as described here.

Optical networks are much better suited for this application as opposed to MPLS and PBT networks which require electronic devices to do the forwarding and label switching. Optical networks can be significantly more energy efficient than electronic networks, but unquestionably far less efficient in terms of multiplexing packets. Tools like Inocybe's Argia can be used to do the traffic engineering of the various optical paths assigned to each address block.

For more information on this architectural concepts please see my blog or presentations at http://green-broadband.blogspot.com