The Network Generation Treadmill - Part 1 The Problem

Posted by adamd at 10:00 AM on Aug 20, 2015


Have you ever noticed that engineers at operators are either busy installing the “next-generation platform” or evaluating “next-generation platforms”? This cycle is tied to the development cycle of monolithic switching platforms such as Broadband Network Gateways. As I have previously blogged, these platforms have a 3+ year development cycle and therefore the “treadmill” has about a 3-4 year periodicity. But how does this impact operators and their customers?

Managing capacity to meet demand & control expenses

Obviously, demand doesn’t mirror the generational steps and therefore capacity planning attempts to figure out the right answer fall somewhere between a high and low projection. In every case these projections are a compromise. Supporting the high projection results in the highest costs and while all customers are happy, the customers that are really happy are the small group of very high demand users, because they are often paying less than they should. Alternatively, the low projection, which is the cheapest makes everybody unhappy because all customers receive unacceptable services. Attempting to draw a line through the middle, (now shown in the graphic) results in reduced costs but accepting that some population of good customers will get unsatisfying service. We have all experienced this oscillation first hand at home if we have had the same broadband provider for more than a couple of years.

So the treadmill continues with each generation promising to solve all of the problems experienced with the last and promising a better price/performance ration. Allow me to digress here — “price performance” is code for “the BNG costs the same or more than last time but but it has much higher capacity so there it is cheaper in relative terms”. Applying this same logic to cars would work something like this. A Lamborghini Huracan cost $245K and has 610HP, and a Nissan Versa costs $14K and has 109HP. Price performance for the Lambo is $416, the Nissan $130. While the example seems a bit farfetched, with a little thought the analogy is not as crazy as it sounds! This price/performance logic does not capture the true cost, because it fails to include the costs of other devices and the change in network topology required to aggregate sufficient traffic to continue to benefit from this higher performance.

Large scale project every time

The “generational treadmill” has further business impacting consequences. Each generational “upgrade” is a significant project because it’s effectively a new system; evaluation, testing and planning are required. The level of effort and complexity usually results in only a sole source supply decision, simply because it’s too costly, complex and time consuming to have multiple sources of supply.

To achieve “price/performance” vendors usually implement new, higher-speed interfaces to their platforms such as the 100GBe. These new interfaces need to connect to something therefore interconnected equipment throughout the network topology also needs to be upgraded to the new, expensive (but better price/performance) interfaces. This generational upgrade, just like the last one, is a complex, expensive multi-year project. The treadmill translates into engineering flip-flopping between preparing for the upgrade and doing the upgrade. Somewhere along the line they are supposed to work on developing new products for customers — but who can find the time?

No getting off the treadmill

There is no getting off this treadmill using the current monolithic hardware platforms. It’s logical to assume that you can run multiple platform generations in the same network, however in practice this is much more difficult. Because these platforms are integrated hardware/software, functions are tied to hardware versions and the addition of a new services (I know, who can find the time?) often forces the need to upgrade to gain feature uniformity. Further managing a piecemeal combination of hardware and software versions with different functionality adds significantly to operational costs. But even if you did undertake the cost and complexity of supporting multiple generations, it would not do that much good. Vendors generally only offer hardware and maintenance support for 7 years, which is about 2 generational cycles. Their reasons for providing 7 years of support are valid but they are caused by the monolithic custom design of their own platforms.

Can Webscale can get you off the treadmill?

The short answer is yes, and the subject of my next blog installment. Webscale datacenters operate in a continuous non-disruptive upgrade cycle with dynamic resource allocation provided by the cloud management software. Server platforms are continuously upgraded; for example if you spent $5K on a server last year and $5K on a server now, you would be getting a faster, more capable platform this year for the same cost. In comparison, if you spent $100K on a BNG line card last year and $100K on a BNG line card this year, you would get the same line card. Software-based systems and cloud management allow resources to be allocated as required to meet demand enabling continuous demand management, the generational treadmill. I’ll explain how this is done in part 2.