Where 10-Gbit/sec Ethernet is going

Oct 1st, 2003
135327

The single most important factor behind the rise in bandwidth requirements is the growing demand imposed by new bandwidth-consuming applications. A typical corporation or business may use a variety of applications, each having different capacity, bit-rate, and transport-speed requirements. For example, while transmission of compressed voice has a low bit rate, the bits need to be available on demand; conversely, file transfers require a high bit rate but can wait for capacity to become available. Applications such as video streaming and voice telephony as well as transaction-based systems like online purchasing and application hosting depend on continuous access to a broadband-network infrastructure. The overall result is that the escalating demand for bandwidth seems to have no end.

The Ethernet networking standard, widely used in LANs, is increasingly seen as an attractive option for achieving end-to-end connectivity across WANs and MANs. Carrier networks that use Ethernet alongside traditional voice transport protocols can reduce the need for conversion between different data transport protocols and standards.

The emergence of 10-Gigabit Ethernet (10-GbE) creates new opportunities for data transport and extends the value and life of Ethernet technology. 10-GbE employs the Ethernet format and provides seamless integration of LANs, MANs, and WANs. It also eliminates the need for routers, which are much slower than switches, because 10-GbE requires no packet fragmentation, reassembling, or address translation. It offers straightforward scalability, with simple upgrade paths, and provides a low-cost solution to the growing demand for bandwidth.

As 10-GbE enters the market and equipment vendors deliver network devices to implement it, multigigabit bandwidth will be combined with intelligent services to create networks with backbone and server connections sporting bandwidths up to 10 Gbits/sec, making applications such as converged voice and data or packetised voice and video running over Ethernet a viable option. 10-GbE technology not only increases the speed of Ethernet to 10 Gbits/sec, but also extends its interconnectivity and operating distance up to 40 km. With links up to 40 km, companies that manage their own LAN environments can strategically choose the locations of their data centers and server farms, allowing them to support multiple campus locations within that 40-km range.

Within data centres, switch-to-switch applications as well as switch-to-server applications can be deployed over more cost-effective short-haul multimode-fibre networks to create 10-GbE backbones that support the continuous growth of bandwidth-consuming applications. With such backbones, companies can easily support the emerging high-bandwidth connectivity in workstations and desktops while reducing network congestion and will be able to implement such bandwidth-intensive applications as streaming video, medical imaging, centralised applications, and high-end graphics.

In the current environment, however, 40-km applications are somewhat uncommon. Rather, the market demand is for reaches of 10 km or less. In fact, the IEEE 802.3 High Speed Study Group had estimated that, at the end of 2000, 88% of the links within the installed "in-building" backbones were less than 300 m in length, with 25% being under 100 m. Likewise, about 56% of the installed fibre in campus backbones had links that were under 500 m.

We believe there is much potential for the cost of 10-Gbit/sec communications to drop significantly, considering the development of 10-GbE-based technologies and their enabling devices. For example, unlike the early 10-Gbit/sec lasers used in telecommunications applications, the 10-GbE technology, as defined in the IEEE 802.3ae specification, can use lower-cost non-cooled optics and vertical-cavity surface-emitting lasers (VCSELs), which can lower device costs.

An Ethernet-optimised infrastructure is beginning to develop in the MAN, which is generally considered to be the worst bottleneck for efficient end-to-end data transfer in the optical transmission network. The technology roadmap of most switch, router, and MAN optical-system vendors, therefore, should include 10-GbE as a way to enable the cost-effective gigabit-level connecting of customer access equipment and service providers and provide simple high-speed, low-cost access to the MAN optical infrastructure.

As 10-GbE begins to be deployed in points of presence, enterprise backbones, and SANs, smaller high-density platforms are required to provide new levels of density, power efficiency, and ease of integration to network equipment OEMs implementing 10-Gbit/sec links. At the heart of these platforms is the optical transmitter/receiver module.

The Figure shows the evolution of 10-GbE transceiver devices as they grew out of multisource agreements (MSAs). The earliest pre-standard implementations used an XSBI interface through 300-pin transponders that required multiple chips and pigtails for optical connection, making it a large bulky solution. That evolved into the XENPAK form factor, which integrated the PHY, eliminated the pigtails, and added "hot-pluggability" digital optical monitoring capabilities. That was the first attempt to define and specify a consistent form factor, connector type, and electrical pin-out based on the IEEE 802.3 XAUI interface for 10-GbE transceivers to fully meet the IEEE 802.3ae standard.

The primary early adopters of this form factor were high-end enterprise switch manufacturers. However, next-generation server-network-adapter manufacturers found XENPAK still to be too unwieldy for their needs, so the XPAK MSA was founded. The latter defines a form factor based on the XENPAK electrical specifications but physically optimised for server adapter cards, which must comply with PCI mechanical requirements. XPAK modules offer higher on-board densities, power efficiency, and ease of integration into networking equipment, making them ideal for both server-adapter and high-port-density switch applications.

To further reduce the package size and increase flexibility, the XFP MSA was founded. XFP defines a new 10-Gbit/sec serial electrical interface (XFI) and moved the PHY back outside the module. That increased flexibility by allowing the optical module to be changed without changing the PHY and by making the transceiver protocol-agnostic. However, it sacrificed power and overall footprint by requiring an external PHY and dual clock and data recovery chips.

There is an increasing demand for the legacy low-bandwidth multimode fibre to support 10-GbE. The current standard that addresses use of this installed fibre infrastructure for 10-GbE is 10GBase LX4. But to implement that requires multiple components and a complex assembly and test process. As a result, the number of companies supporting it is declining.

An alternative solution, which we believe more likely to be successful, is the electrical dispersion compensation standard now being developed and used in its initial form by some polarisation-mode dispersion suppliers. This solution uses the current 10-Gbit/sec serial optic technologies.

More in Transmission