spirent.com

Metrics for Network Performance

By Chris Chapman On December 20, 2011
Networks
No tags assigned.

Joint post by Gail Ferreira & Chris Chapman

What do you look at to size a network solution, to make sure that it has adequate capacity for your network? What measurements should you focus on when evaluating the capacity of a component in a network?

1) Gbps of throughput?  But how much traffic is being dropped in the process of reaching peak Gbps of throughput?
2) Connections per second?  Will there be enough CPS to cover peak loads?
3) Or maybe total number of connections that enable all users to be supported?
4) Or is CPU utilization under load the most important indicator so that you can cover peak loads?
5) Or maybe it’s 5 9’s of availability and a failover pair is more important than capacity?

There is an entire constellation of metrics, but which one is the most important?

 constellation

Do you measure and evaluate each specification (e.g. throughput, connections per second, max concurrent, etc.) separately and exclusive of each other? If so, how do you then apply these technical metrics back to a business metric directly, such as the number of subscribers a mobile operator needs to support on their networks?

It’s not easy to evaluate current capacity and meet anticipated future capacity needs, especially as security threats continue to increase. The pitfalls include:

  • Focusing just on a single metric
    One might anticipate that throughput is the single most important metric, but then find that the traffic profile associated with the throughput of the current or anticipated environment isn’t comparable to the throughput claims on the devices chosen.
  • Selecting a few elements to measure
    Maybe throughput and connections per second are key, but focusing only on them could limit the ability to scale to the total connections needed.
  • Focus across several capacity measures
    But lack an accurate characterization of current and projected traffic to which to contrast alternative devices.
  • Fail to realize the impact on the user experience
    A device under load may not provide a satisfactory user experience even if it can pass all traffic.

How essential is it to look at so many variables?  Is it really necessary to have a “wide-angle” perspective of this constellation of metrics?  If that were the case, it could mean that there is no substitute for actually measuring the anticipated workload.  Actually measuring and comparing the real-world capabilities of devices eliminates the guesswork in finding the best solution for your network’s needs.  That avoids the problem of trying to adapt to inadequate devices on your networks. For instance, reducing security to obtain the desired performance, or worse, turning off security altogether, as discussed in the recent survey results. 

Gail Ferreira is the Product Marketing Manager – Mobile at Crossbeam
Chris Chapman is Technical Marketing Engineer at Spirent Communications

 
comments powered by Disqus
× Spirent.com uses cookies to enhance and streamline your experience. By continuing to browse our site, you are agreeing to the use of cookies.