100G Ethernet and beyond: preparing for the exabyte Internet
by Raza Rizvi
July 2009
Executive Summary
This is a time when a wave of new networking technology will enter the market
It is now approaching 30 years since the Ethernet standard was first published. Since then it has become the dominant mechanism for communication between devices at the data link layer of the OSI networking model, increasing in speed from the initial 10 megabits per second standard through 100 megabits and 1gigabit per second, to 10 gigabits per second (Gb/s) today.
Over the last decade, the availability of, and the demand for, information has increased at an unprecedented rate and this is driving the need for increases in the access speed across networks, and between networks and servers. Network managers are seeing a new scenario materialise, one that has moved away from a period of predictable traffic growth where capacity planning could be applied to both the Ethernet network and the separate networks used for voice, video, and storage. In the past, end-user access to the network was at a speed many multiples lower than the speed supported in the core backbone and the backbone itself was a benign transporter of packets. Now we see aggregation of multiple input data types (voice, video and storage) onto a single homogenous Ethernet network with fluctuating traffic demands, where the client access nodes or networks are connecting at speeds approaching that of the backbone. The backbone is required to provide services that are more reliable and predictable, for longer periods, and to be able to cope with rapid increases in demand with no ebb. The standards bodies, particularly the Institute of Electrical and Electronics Engineers (IEEE), have been pushed along behind the curve of this wave for more bandwidth and have steadily improved both the speed and the simplicity of deployment of Ethernet products. To date, Ethernet has evolved in factor-of-ten increases, but some argue that the next logical step – to 100Gb/s – is a bridge too far and that 40Gb/s is a more realistic goal.
It is in this state of flux that network managers will be required to make decisions about the next generation of networking equipment. Developments in Ethernet technology raise questions about the appropriate response from institutions with regard to new equipment. Larger universities, major research centres and institutions are likely to be the first to face these questions and there may be very considerable investment at stake: a typical 224-port switch with 10G on every port is of the order of £250,000. Newer equipment with 40G or 100G is likely to be more and early adopters are often financially ‘penalised’. Taken together with the buying cycles of academia, network managers have to look ahead to what the demands on the network are likely to be to support user applications. Capacity planning must consider the likely levels of traffic in the distribution and core network equipment over the next 3–5 years, allowing for depreciation, in accounting terms, and suitable levels of return on investment.
This report explains why there does not appear to be a consensus for a single target and looks at the implications that may have for network managers in HE, particularly in light of JANET's recent announcement of trials into 100Gb/s transmission. It will review some of the technical implications of a move to 40Gb/s or 100Gb/s and make recommendations for how to maximise purchasing decisions at a time of flux in the industry. Finally, it will look ahead to the development of terabit Ethernet in order to put the continuing evolution of Ethernet into a longer-term context.