Networking opportunities: Beyond 40Gbps

With OC-768 now being deployed and 10GigE going mainstream, many network managers, engineers, entrepreneurs and investors alike have begun to consider what the next generation of networks will look like. There are already signs that power users such as Internet exchanges, very large data centers, and high-performance computing installations are straining under 10GigE’s limits and are using 10GigE LAGs (link aggregations) at Internet exchanges and POPs. In the public networks, and to some extent in enterprise networks, the primary factor driving bandwidth demand is, yes you guessed it, video! For anyone working in telecom in the last 15 years, that is an old and rather familiar refrain but for the first time it actually seems to be happening due to IPTV and other content rich applications. Mobile video is also another possible factor but as of now remains undeveloped.



It should be stressed that the use of 10Gbps LAGs is very much at the leading edge for even if everyone who would like to replace their 10GigE LAGs with (say) 100GigE were actually able to do so, we would likely be talking about a market of no more than a few hundred 100GigE ports. What’s more, the process for defining the next generation of Ethernet at the IEEE has barely begun and for the next generation of SONET/SDH it has not begun at all – and indeed may not. Optimistically, Ethernet standards will be finalized in 2009 or 2010 with the first group of products appearing right before that time. (Although, a few companies such as Infinera and Luxtera have combined WDM technology and optical integration to support very high data rates today.) While nothing has been settled regarding data rates they most likely will be either 80, 100 or 120Gbps. (There is a strong possibility that by the time standards emerge 80Gbps might be too late.) 160Gbps appears unlikely due to limitations on available technology. Regardless though, the next generation of networks are going to require some significant engineering at both the component and networking level. Issues such as power consumption, processing speeds, thermal efficiencies and even materials will have to be dealt with in both the electronics and optical areas. And, with new types of multimode fiber (MMF), or single-mode fiber (SMF) being required for the next generation networks, dealing with existing copper and older MMF cabling infrastructures will need to be addressed.



Some of these challenges, at least initially can be ameliorated by using parallel solutions. For example, for 100GigE, it has been proposed that 10 x 10Gbps, 4 x 25Gbps or 5 x 20Gbps formats could be used which would enable some existing technology and components to be used, especially the 10 x 10Gbps format. The parallel approach has been used before in early 10 GigE formats – in CX4 (for copper) and

LX4 (for fiber) -- and will almost certainly be the route that is taken with the earliest “next gen” networks. These parallel approaches introduce problems of their own though. Almost everyone we have talked to would prefer a serial solution but unfortunately few see serial transmission as a practical solution for many years to come. Our sense is that the choice of parallel scheme is shaping up to be a somewhat controversial and important issue. We have summarized how we see the “balance of power” in such matters below.



Regardless of the problems inherent in developing what’s next in networking technologies, the bottom line is that these new networks are the future opportunities for processor, component, module and networking companies. Furthermore, we say “good luck” to anyone trying to get into the conventional 10Gbps market. There are already too many firms selling 10GigE modules and we expect that market to contract over the next few years. The opportunities will therefore have to come from pre-standard next-generation network products that cater to the needs of market niches and/or new kinds of technology whether those be integrated optics, new types of dispersion compensation technology, improved modulators or something else deemed essential for operating future networks.



Nonetheless, some of the new technologies targeting future networks may actually spawn more cost effective 10Gbps products that will open up more cost sensitive markets for 10GigE and possibly even OC-192.



Much of the technology work over the next few years will focus on developing transmission components that are suitable for the various parallel WDM-based schemes currently being proposed. There is a certain urgency about this work for unless there are some real signs of progress in the next year to two years it will be almost impossible for standards groups to adopt 20Gbps or 25Gbps as an underlying technology. A key assumption of such work (quite reasonably) is that standards will not be adopted unless the technology available to support it exists. Today, nobody really knows for sure if you will be able to cost effectively directly modulate a DFB laser at 20Gbps and many people believe that 25Gbps cannot be done in the time frame necessary to meet standardization timetables. The failure to develop a robust yet economical 20Gbps transmission technology would almost certainly hand an early victory to a 10 x 10Gbps schema.



In addition to parallel schemes using CWDM, another form of parallelism that may be deployed is parallel optics using 12 x 10Gbps over ribbon fiber. Ribbon fiber is not well liked by network managers and this approach would probably be useful only over a few meters. Nonetheless, parallel optics is already quite widely used in computing centers with the current commercial leading edge being 12 x 2.5Gbps. The OIF has defined parallel optics at 10Gbps and the 10Gbps VCSEL arrays that will be necessary for 120Gbps parallel optics are quite close to commercialization.



Challenges and opportunities will also abound in electronics as well as optics. New higher speed MACs will be required and some folks propose that this will necessitate new designs including some kind of distributed functionality across multiple identical chips, possibly with aggregation at the physical layer. In general, switching speeds will have to be achieved that match the data rates of the networks, which, in turn, will certainly push the envelope in terms of the processes and materials being used. Yet, at the same time the electronics will need to be manufactured with standard processing equipment, use commonly available materials and be designed using existing CAD systems. The 65nm or even 45nm node production technology with some use of materials other than silicon will also be required. InP, InGaAs, GaAs, InGaP and SiGe have all been suggested in this context. Some engineers, for example, believe that SiGe would be needed for SERDES; others say that InP would be needed for the electronic mux/demux and perhaps for certain drivers. Finally, it seems highly likely that EDC will play an even bigger role than it is beginning to do in the 10 GigE market, although the new 100Gbps will be more complex than those we see today.



While standardized (not to mention volume level) products that operate at 40Gbps are almost certain to be several years away, CIR believes that even the tentative steps that are now being taken to define what’s next in Ethernet and SONET/SDH are likely to have much more immediate impact on networking and component businesses than such timeframes would seem to suggest. Early technology development must start now if the standardization efforts are to meet their forecast timetable. This development process is also likely to push the cost curve in getting today’s 10Gbps and 40Gbps technology down to more attractive price points.