10 Gigabit Ethernet over copper wiring has potential in data center
New Data Center Strategies Newsletter By Andreas M. Antonopoulos, Network World, 08/15/06
Data centers underwent a network convergence in the early and mid-'90s as Ethernet became the de facto standard. But as new applications created a need for high-speed and low-latency network fabrics, the data center networks became more fragmented. In today’s data center you might find 100Base-T Ethernet, Gigabit Ethernet over copper or fiber, 10 Gigabit Ethernet over fiber, Fibre Channel for storage, InfiniBand for server interconnects, and so forth.
All of these different network types require specialized skills, different media, and various types of smart switches to interconnect them. The end result is increased cost, increased complexity and limited flexibility. If we are to realize the flexible data center of the future, where servers and storage are pooled and made available to applications on demand, it will be helpful to standardize the networking fabrics and protocols.
With every generation of LAN technology, I’ve always heard reports of the imminent demise of Ethernet. At every new wave of innovation, some seemingly impassable barrier stood in the way of higher capacity. Yet Ethernet always seems to re-invent itself, and any bets against it seem to be short-lived. Granted, there are always going to be some special circumstances where a non-standard or highly specialized fabric is more appropriate. But in general, standardizing on a few fabrics, link-layer protocols and media types can make data center networks easier to manage and more flexible.
A few recent developments foretell a re-convergence on Ethernet in the near future. A working group in the IEEE (802.3an) has been working on 10 Gigabit Ethernet over twisted pair (shielded or unshielded, Category 6 or 7). Recently, some vendors have started demonstrating prototypes which remarkably work over even Category-5 unshielded twisted pair (UTP), albeit at distances not exceeding 45 meters. In the average data center, 45m is more than enough to replace a great deal of proprietary server-interconnect fabrics for applications in the high-performance computing space.
While there are still power consumption issues and the inevitable price issues, 10GBase-T could revolutionize data center networking. It could now be possible to build a data center with Cat-6 UTP cabling throughout, supporting 10M, 1G and 10G Ethernet on a single medium and plug type (RJ-45).
Companies could conceivably consolidate InfiniBand and Fibre Channel-based networks, replacing them with 10GBase-T. This could lead to rapid adoption of iSCSI and even more adoption of blade server systems.
The data center of the future could be based on a single cabling format (Cat-6 or Cat-7 UTP) and link-level protocol (10GBase-T) providing for all types of networking:
* Server-to-client connections
* Data center-to-Internet connections
* Server-to-server interconnect replacing InfiniBand
* Storage-to-server interconnect replacing Fibre Channel
* Data center-to-data center interconnect over metro Ethernet
It’s a single set of skills, plugs and protocols to rule them all. Perhaps it’s time to create the second law of Metcalfe: Never bet against Ethernet.