Running Networks Hot: Urban Myth or Viable Strategy?

In my previous blog, I mentioned the trend of running networks hotter in order to squeeze more performance out of the existing assets, delaying capacity upgrades until absolutely necessary. This discussion is not new; it is revisited every time the economy slows down. Network traffic keeps growing in spite of — or perhaps because of — the downturns, and while increasing the utilization level of the network certainly can save costs in the short term, it is not without risk, as service quality and reliability ─ and hence customer experience, might be impacted.

The first thing one might think about when looking at the possibility of running a network hot is to use capacity management tools to predict traffic flows and to tweak the network in times of stress. While this may appear to be a viable solution, the reality is that the vast majority of traffic management tools come up short in their capabilities, especially as networks get increasingly complex.

Not so long ago, running networks hot was not a prime concern of network administrators. Large-scale networks were built primarily to carry predictable and pre-provisioned voice and other deterministic circuit-oriented traffic, while the more dynamic packet-oriented data networks (frame relay, ATM, IP) were much smaller and traffic was limited and easily managed. Things have changed since then!

Today’s networks are highly dynamic and in constant state of flux. Carriers offer voice and data services, as well as video, wireless, gaming, and managed networks, frequently all on common equipment and increasingly on the same network. These complex networks have varying data patterns in addition to frequently changing services that need to be delivered to shifting endpoints. It is almost impossible to keep close track of so many moving parameters. Furthermore, as networks become more decentralized and with users at times capable of determining the routes of their traffic, capacity management may in some cases decrease the performance of a network (see Capacity Management in Decentralized Networks by Yasushi Masuda and Seungjin Whang).

There is one tool that service providers have at their disposal to run their networks hotter, while still avoiding congestion: bandwidth throttling. While many have announced they are considering it, some have already placed caps on bandwidth usage, much to the dismay of consumer rights groups and content providers. In most cases, only the heaviest users are hit at peak times, as service providers try to cope with increasing demand across their network. However, if bandwidth usage continues to grow at its current pace, it could outpace new infrastructure development in just a few years, and we can expect bandwidth throttling to become mainstream policy. This is a very real and serious problem that needs to be addressed. As early as 2007, Nemertes Research warned that the information superhighway could become clogged with data by 2010.

We are witnessing the same in the wireless access networks. An article in the Sept. 22, 2008 issue of Business Week reports that wireless companies are effectively struggling with data demand. As mobile users are increasing their use of data-intensive applications, they are switching away from plans that charge on a per-KB basis to unlimited data plans. In fact, comScore reports that between June 2007 and June 2008, there was a 58 percent increase in the number of users who subscribed to unlimited plans. While currently only 2 to 5 percent of users are affected by upper limits on “unlimited” data plans, Craig Mathias, founder of consultancy Farpoint Group, estimates that in the next year that number could double.

Prior to bandwidth throttling, service providers mostly resorted to oversubscription to increase network utilization. The real question is “how much do you oversubscribe?” The rule of thumb used to be 4:1, but it is not uncommon anymore to see 12:1, 20:1, or even higher as companies try to stretch resources to their limits. The obvious downside is that you run the risk that all users want to use their bandwidth concurrently. However, the risk is low only if oversubscription is low. Very high oversubscription ratios combined with unforeseen events that require rerouting could lead to catastrophic events leaving many customers without service, at a very high cost. Estimates of lost revenue because of network unavailability range from $2,000 per minute up to several million dollars per hour, with the impact being compounded as time increases (Forrester study, Fiber Channel Association). With the stringent SLAs service providers have to agree to today, it only takes a single event like this to wipe out any savings realized from deploying less equipment and pushing oversubscription ratios too high.

In all likelihood, service providers will run their networks a little hotter in some places –after all, everyone is trying to do more with less. But there are some networks, services and applications for which it simply isn’t an option. And even if parts of the network can run at higher utilization, there really isn’t a magic bullet on how to do it. It takes a combination of smart network planning, the right level of oversubscription and most likely some form of bandwidth throttling or traffic shaping. And of course, a little bit of foresight of the future. Crystal ball, anyone?

Luc Ceuppens is vice president of product marketing, High-End Systems Business unit at Juniper Networks(JNPR).

Share this article: Email, Slashdot, Digg, Del.icio.us, Yahoo!MyWeb, Windows Live Favorites, Furl
Add this article feed to: RSS, My Yahoo, Newsgator, Bloglines