25 G vs. 40 GbE Comparisons for the Enterprise

Choosing between 25 GbE and 40 GbE isn't simple or easy, but knowing some key differences will help.

By Arthur Cole | Posted Feb 5, 2016
Page of   |  Back to Page 1
Print ArticleEmail Article
  • Share on Facebook
  • Share on Twitter
  • Share on LinkedIn

Bandwidth considerations in the enterprise are diverging across two distinct development paths, and while they are not mutually exclusive— in fact, they lead to the same destination—they nonetheless throw a wrench into what would otherwise be a fairly straightforward progression from today’s network infrastructure to the broader, more dynamic environments of the future.

For a while, it looked certain that the powers that be were content to proceed from 10 GbE to 40 GbE and then on to 100 GbE. That was all well and good until hyperscale infrastructure started to take hold. The top players in that field prefer a 25 GbE standard, which can then simply be doubled and then doubled again to 100 G. As The Next Platform’s Timothy Prickett Morgan points out, things can get a little expensive in scale-out infrastructure when building either 40 or 100 gig architectures based on 10 GbE lanes. The math works out much better when you start with a 25 gig lane. As it stands, however, most networking vendors are keeping their hands in both the 40 and 25 gig pies before determining which way the market breaks.

Demand for 25 GbE will likely center on the desire to drive down networking costs between the server and the ToR switch, says Tolly Group founder Kevin Tolly. As he explained to Biz Tech recently, costs for 10 GbE will continue to drop, which means enterprise will likely use it more frequently for direct server links. This will inevitably produce bottlenecks between switches if they too stay on a 10 GbE footing. Jumping to 25 G on the uplink will allow organizations to push 10 G in the rack without oversubscribing the switch infrastructure. Sure, you could do this with 40 G as well, but it will cost more and won’t be as power efficient.

In a way, the decision between 25 and 40 G has to do with forging links to the past or to the future. As mentioned above, 40 G supports four lanes of 10 GbE cleanly, while conversion to 25 G means you up the bandwidth of each lane right away. A 40 G solution, then, will require a slightly more complex conversion to 100 G when the time comes versus a 25/50 solution. So one of the key questions network architects should be asking is whether they want to make things easier for themselves now or later.

But that question might not be as simple as it sounds, because sub-10 GbE standards are starting to crop up around the enterprise that may make the decision to go with 25 G a little easier. According to Dell’Oro Group, 2.5 and 5.0 Gbps formats are taking hold on campus networks and emerging wireless infrastructure, and these could very well end up supplanting the data center as workloads gravitate toward “new economy” initiatives like Big Data and the Internet of Things. As the group’s Alan Weckel notes, many organizations are swapping out 1 GbE infrastructure for 2.5/5 Gbps, if only to push security and other functions to the network edge.

This also starts to make sense when you consider that even upgrading to 10 GbE requires a costly retrofit of network infrastructure to CAT 6e cabling. Many organizations are looking to avoid, or at least delay, this prospect through new specs like NBase-T, particularly as wireless traffic starts to hit the enterprise. NBase-T allows for 2.5 and 5 Gbps bandwidth across common twisted pair wiring up to about 100 meters. Wiring is usually the longest-serving element on the physical layer, for obvious reasons, so anything the enterprise can do to accommodate wireless traffic without pulling entire buildings apart is welcome. But it also puts a good chunk of the data center network on a 2.5 Gbps footing, which only makes it easier to adopt 25 G, and ultimately 100 G, when the time comes.

Unfortunately, there is no way to avoid the bandwidth crunch that is on its way. As the world becomes more connected, data productivity will be increasingly defined by how quickly workloads can be transferred or distributed across increasingly distributed infrastructure.

Decisions regarding the exact approach to easing bandwidth constraints will be as diverse as enterprise infrastructure is today, however, so don’t look for quick and easy answers. Ultimately, it will come down to where you are today and where you hope to be in the future.

Comment and Contribute
(Maximum characters: 1200). You have
characters left.
Get the Latest Scoop with Enterprise Networking Planet Newsletter