Purchasing decisions in networking have long been dominated, especially on the non-provider side, by CapEx. At one point, Cisco had commissioned a study from IDC that indicated just under two-thirds of the purchasing decision was based on capital costs. And if you look at the rise of merchant silicon and white box switching, you would have to conclude that the industry has been in a years-long cycle to address these concerns.
In fact, several of the major industry announcements in 2013 were focused primarily on solving the CapEx problem. Arista has long been espousing the benefits of merchant silicon in driving the price per port down, and they continued that drumbeat with their Spline announcement in a move back to middle-of-row switching. It would seem that the rest of the industry has caught up as well. One of the more subtle aspects of Cisco’s Insieme announcement was their merchant+ strategy, which will bring their hardware costs down below Arista’s.
None of this is new, but it does foreshadow a shift away from CapEx as the primary decision criterion. If everyone converges on the same narrow set of merchant silicon, the prices will become almost indistinguishable. When price is not the main reason to make a decision, people will rightfully shift their focus to longer-term operational costs. This intersects nicely with trends like SDN, NFV, DevOps, and network virtualization, all of which address various workflow aspects of managing a network.
This year should be the year where OpEx gets more play. I know that network management has claimed OpEx benefits for years, so it might not necessarily play out in terms of great vendor focus, but it will certainly start to shape customer decisions more than it has in the past.
But if the march towards lower CapEx has taught us nothing, it should be that much of the industry will make the move together. Sure, there will be different means of achieving lower OpEx: Cisco will integrate their solutions, Arista will lean towards on-box automation, and the SDN companies will elevate their controller strategies. Customers will absolutely need to evaluate how OpEx reductions will impact them, assessing which PowerPoint slides are likely to lead to reality and which make for great presentations but are unlikely to yield tangible results.
Ultimately, though, it won’t be too long before even the OpEx discussion becomes a supporting point.
It simply cannot be that the end game in all of this disruption is to create the exact same networks as today but at a lower price. The energy around new technologies is not purely price-oriented. Movements like SDN are driven by a need for the network to fundamentally do more than it is doing today. Connectivity is table stakes, and table stakes rarely act as a long-term differentiator.
Put differently, is the CIO going to be evaluated primarily on budget over the next five years?
It seems more likely that, having gotten cost under control, CIOs will be asked to demonstrate their impact on the rest of the business. That impact goes well beyond budgets, and it means that forward-looking CIOs should be thinking now how their roles intersect the company’s broader strategic initiatives.
For example, for large, distributed companies, the major measure of success might be productivity and end-user satisfaction. Do the lines of business have access to the applications they need when they need them? Is productivity more or less a constant regardless of whether the worker is in corporate headquarters, a Starbucks down the street, or somewhere in India or China?
So what does it take to make these types of experience guarantees?
First, IT needs to know how to quantify experience. I have talked about this before so will spare you the verbiage here. But once you have the experience SLAs defined, you need to have the infrastructure instrumented so you can determine how you are performing against them. Interestingly, application experience is an aggregation of compute, storage, networking, and even the application. Accordingly, instrumentation needs to span the various elements.
But simply knowing whether you are meeting an SLA or not doesn’t mean you are delivering expected results; it only allows you to report on how you are doing. Eventually, you need to feed that instrumentation back into the infrastructure so that you can make adjustments.
In this context, though, tuning goes beyond optimization. The objective here is not to give every bit of traffic an optimal path. Architectures ought to be doing that today. If you think about it, most infrastructures are built agnostic of the applications that run on top of them. Without any information, the best they can do is get traffic from A to B in the fastest (or most secure) way possible. Sure, there are QOS policies that dictate performance to some extent, but the environment is very static. The point is that if your infrastructure is somehow not already delivering the best possible performance in aggregate, you should be asking why.
In a world that is built around application experience, however, the objective is slightly different. Assume that performance is directly related to some capacity (CPU cycles, storage space, network capacity, and so on). If you want to tune the infrastructure based on SLAs, it means you have to allocate resources differently. The real detail here is that a shift in resources to benefit one application will come at the expense of something else (unless you are over-provisioned across the board, in which case I would argue that you haven’t dealt with your cost issues yet).
The magic here is that CIOs will need to have a more explicit idea of which applications are the most important. Treating all applications the same prohibits you from making meaningful adjustments.
So the question for people building out infrastructure today is one of strategic prioritization. Is your plan to deliver AppEx tied to an over-provisioning strategy? And if it is not, have you gone through the preparatory steps of identifying those applications that are most critical to your company’s mission?
I expect AppEx to take time to rise to a level of prominence, but there are absolutely things people can do now to prepare for what will eventually prove inevitable.[Today’s fun fact: Cats can hear ultrasound. This makes pregnancy particularly obnoxious for cats with overzealous health plans.]