Why you don't need to throw bandwidth at performance problems

You don't need to throw bandwidth at performance problems
Dealing with performance issues more shrewdly

CIOs say that one of the dilemmas they face when implementing services such as VoIP is how to provide an effective service without throwing bandwidth at the problem. There are two problems with doing this: it takes time, and it's expensive.

New circuits can take anywhere from 30 to 90 days to deploy once the order is placed, and every organisation is working with limited budgets. The cost of bandwidth is a particular challenge in emerging markets such as the Middle East.

So it is important to be absolutely sure that any increase in bandwidth is needed and can be deployed in the right place before going ahead. In many cases, it is possible to improve the end-user experience and reduce operational costs by simply using existing bandwidth more efficiently.

To make an informed decision, organisations need to consider:

  • Which network links are most critical?
  • Is bandwidth being consumed for business purposes?
  • Can I downsize a link while maintaining business service quality?
  • How can I demonstrate that an increase in capacity is warranted?

Managing Network Capacity

There are three common approaches to managing network capacity, but all have their limitations. Traffic usage totals only show congestion in extreme cases, while peak utilisation shows the days in a month that had a busy minute but does not give insight into the amount of time a link is congested.

Taking a long term view of average utilisation is more informative, but traffic spikes and even brief periods of congestion are hidden by the highly aggregated averages. Truly understanding network utilisation requires a high degree of granularity, but obtaining this fine level of granularity over a month or a year requires a vast amount of data to be stored and displayed.

A better approach is to use burst utilisation, which indicates the amount of time interface utilisation is greater than specified thresholds. This enables the network manager to determine how long the congestion of a link exceeded 80 percent utilization or any other chosen threshold.

They can then decide whether to upsize (or downsize) a link or investigate how the link is being used. The advantage of burst utilisation is that link congestion levels can reported based on 1-minute granularity regardless of the reported time frame (day, month, or year) without loss of information fidelity.

Network managers typically want to begin to keep an eye on a particular interface when it spends more than 10 percent of time above 80 percent utilisation. This translates to a little more than a half day out of a typical working week.

When the utilisation burst reaches 20 percent time spent at the 80 percent threshold i.e. a full working day then it may be time to either upgrade the link or investigate how it is being used.

  • Roger Holder is Fluke Networks field marketing manager for network and application performance solutions for the enterprise. He has 20 years' experience helping organisations design and manage the performance of mission critical business applications.