Despite the rapid rise of cloud computing expected over the next decade, most enterprises will continue to invest in internal physical infrastructure as well. But with the ROI of the cloud hanging over all future decisions, the pressure is on to derive the greatest benefit with the least amount of capital.
That's why network convergence is such a big deal. The Linley Group sees convergence hitting high gear over the next few years, which in turn is expected to drive 10 GbE NIC and LoM shipments past the 16 million port range. As long as the vendor community can accommodate the various protocols that are needed for convergence, namely TPC, iSCSI and FCoE, not to mention new virtual environments like VMware and Xen, most enterprises should be well on the way to a single network architecture by mid-decade.
And yet, there is a growing chorus of voices who argue that there is something missing here. It's not that convergence isn't a worthy goal, but more that 10 GbE, on its own, is not necessarily the answer.
For one thing, argues SymQuest Group's Matt Prigge, the continued focus on wire rates alone to gauge Ethernet's suitability as a network fabric is wrong-headed. (Note: The full article requires registration). The fact is Ethernet and Fibre Channel alone have entirely different feature sets, so just because you can carry one over the other does not necessarily mean you'll have optimal performance. In fact, there may be numerous instances in which multiple 1 GbE links would do better than 10 GbE.
As convergence moves forward, echoes IT consultant Stephen Foskett, many enterprises may find that Ethernet suffers from a Goldilocks problem: either too little or too much throughput for a given environment, but rarely just right. Many enterprises have already found a comfortable middle with 8 Gbps Fibre Channel, but are balking at the cost of both server and network retrofits to carry it over Ethernet.
Keep in mind, though, that the ultimate goal here is not exactly "convergence" but the elimination of multiple network tiers, says ComputerWorld's Jim Duffy. Simply dropping 10 GbE into the server/switch architecture won't be terribly effective without the removal of the Spanning Tree protocol, particularly if your goal is to foster a more dynamic, flexible virtual environment. A non-blocking architecture will also allow you to connect servers and switches directly to the core network, skipping the aggregation layer altogether, and removing much of the latency that currently inhibits data flow. Though 10 GbE is still a valuable component in future networks, it needs to be supplemented by developments like TRILL (Transparent Interconnection of Lots of Links) and VEPA (Virtual Ethernet Port Aggregation) if all the promises of convergence are to be realized.
At this point, it is very difficult to draw firm conclusions about the future of the network considering the technology is changing so quickly. Cost/benefit calculations will surely be altered by the advent of increased bandwidth solutions like 40/100 GbE and 16+ Gbps Fibre Channel.
Convergence is a complex problem, and it won't be accomplished through any one technology.