Higher-speed Ethernet (HSE) looms large on the horizon as the IEEE deliver two new network speeds, 40 and 100 Gigabit Ethernet (GbE). However, the arrival of  the HSE solutions so soon after completion of 10GbE standards presents network managers with a dilemma.

To future proof for HSE do they now provision with 8, 20 or just 2 fibres per link, should they use multi-mode or single-mode fibre, and can advances in Cu cable ever be discounted? Ken Hodge, R&D manager at Brand-Rex, assesses the migratory options and advises to proceed with extreme caution.

With the ink barely dry on the 10GBASE-T standards, the IEEE 802.3ba working group has introduced a set of new application standards delivering 40GbE and 100GbE performance in the LAN. Demand for higher Ethernet speeds has come from a number of interested parties, including operators and owners of metropolitan core networks, data centres, supercomputers and corporate backbones.

Commercial 40GbE and 100GbE products are expected to appear on the market in reasonable numbers from 2012/2013. However, organisations will start to demand 40GbE speeds at the equipment points (servers) in data centres by next year, with the connectivity requirement rising to 100GbE by 2017. That said, it is anticipated that 40GbE will account for just a few percent of the market at server level by 2013, while 100GbE speeds will not hit server level for at least 6-7 years – and even then, only in niche, high-performance computing and networking applications.

Meanwhile, backbone, campus and metro links are likely to migrate straight to 100GbE. Some organisations might install 40GbE due to the cost benefits, albeit in much lower volumes initially. Deployments of 40GbE switches could begin this year, and fibre will be required to enable the migration. Both multi-mode and single-mode solutions are envisaged and, in backbone applications or any longer link lengths, fibre will be the only choice.

The key challenge for network managers is going to be in future-proofing the physical infrastructure (PHY) in support of these higher-speed Ethernet technologies. Certainly, the volume of fibre (and thus the capital outlay) required to support higher speed equipment could increase significantly – by a factor of four to support 40GbE, and by a factor of ten to support 100GbE.

The alternative is either to install single mode fibre from the outset and leave it dark until higher-speed (and more expensive) equipment is deployed, or to ensure that there is enough space left in trays and pathways to add cable as upgrades are implemented. In backbone applications, blown fibre provides a viable option.

Negotiating the PHY minefield
The new IEEE 802.3ba standards incorporate PHY based on current VCSEL (Vertical Cavity Surface Emitting Laser) and laser technologies. And since it is not yet possible to modulate single lasers at speeds to generate 100Mbps (other than in laboratories), multiplexing is required. Either wave division multiplex technology (WDM) with standard lasers, or multi-fibre parallel array links with VCSELs, will be used. This means that the cables required will be duplex single-mode fibre, or multi-fibre multi-mode links respectively.

A PHY system for multi-mode fibres using either a 4-lane, 8-fibre (40GbE), or 10-lane, 20-fibre (100GbE) topology is envisaged.In terms of transponder costs, the 4-lane approach will see an estimated two-fold increase on current CapEx for Ethernet-related media, with the 10-lane design introducing a four-fold increase. This is just about in line with ‘ten-times the speed for three-times the cost’ target of IEEE. The PHY systems for single-mode fibres and 4 standard lasers with WDM will cost more than the multi-mode solutions.

Within the 40 and 100GbE solutions, there are four alternative cabling choices, each aimed at providing an optimum link-length deployment. However, there are grey areas, for example, on shorter backbone links the choice is either: duplex single-mode fibre with 4 x 10GbE lasers or 4 x 25GbE laser devices; or multi-mode fibre (for which MPO connectors are required), with 4 x 10GbE VCSELs, or 10 x 10GbE VCSELs.

In terms of cabling, either OM3 or OM4 multi-mode fibres can be used (OM4 offers a slightly longer reach), with MPO connectors or duplex single-mode fibre with LC connectors. There are also new solutions for very short length (<7m), rack-to-rack links, which are copper-based and use small form-factor infiniband-style connectors, much like the 10GbE CX4 technology today.

Further muddying the 40 & 100GbE waters, an alternative solution that employs standard 40Gbps, telco-grade lasers, is now under development in IEEE. Although this technology is already implemented within telecom networks and uses a high-performance laser on duplex single-mode fibres, it still requires further development to work in the Ethernet world.

There is also growing interest in a proposed 40GBASE-T standard in IEEE. If this were to be formalised, it would offer a significantly lower-cost option to the fibre solutions available – even the lowest-cost 40GBASE-SR4 approach (8-fibre OM3 cable). The upshot is that there are not only eight potential 40/100GbE solutions, but a further two if OM4 cabling is employed (for a greater reach than OM3), while there are at least another two technologies in the pipeline.

Ethernet’s history lesson
Further complications arise in planning for 40 & 100GbE in terms of the additional fibre required to sustain current server densities, and whether a top of rack, end of row, or centralised switching architecture should be specified. Centralised switching lends itself well to data centre applications, particularly where operational costs must be optimised, because of the high port costs of switches. This would mean a structured cabling architecture with passive cross-connect will be preferred. Moreover, it would be unwise to bet against higher-speed copper solutions emerging by the time 40/100GbE deployments gain volume.

The key advantages of Cu systems are that they are more cost effective than optics, so a 40GBASE-T would be preferred for short-length links and for installations that are connector intensive (e.g. in data centres). At server equipment level, and certainly in the near term for high volume applications, Cu (10GBaseT) will continue to be the solution of choice.
Not to be outdone, the multi mode development team within the IEEE have recognised that performance is restricted to 100 metres on OM3 fibre optic cabling, and are therefore actively working to extend that distance to a practical length. With OM4, it looks likely that 150 metres will be supported as well.

If Ethernet history has taught us anything, it is that having such a broad array of solid technology solutions available makes it simply impossible to settle on any one option. Network managers will likely have to run with a number of these, and specify according to application. For example, single mode for backbones, campus runs, and metro links; and multi-mode fibre – either 4 x 10GbE, or 10 x 10GbE – for shorter runs. This is significant because the latter choice means providing either an 8-fibre or 20-fibre cable for what was originally intended as a single duplex link.

The alternative will be to wait for the new developing 40GbE serial laser device solution (40GBase-FR4), which will need only 2 single-mode fibres, or the ‘40GBASE-T’ Cu cabling.

Currently, OM3 performance cables using 10GBASE-SR devices is the most economical solution for short run length optical links in buildings, with OS1 or OS2 cabling and more expensive devices are used on any lengths greater than 300m. At the same time, 10GBASE-T and ClassEA cabling are the optimal Cu solution for the data centre, short length backbone or office networks.

Whilst it is too early to advise on the preferred cable solutions for future proofing for 40/100GbE, if we were to speculate, the media required per installation scenario would be as given in Table 1. When future-proofing for longer-length network links, we would definitely advise that provision for single-mode fibre was made. The indecision in the industry, as ever, surrounds choice of cabling to support shorter distance links.