Most networks now have a degree of subnetting, VLANs, routing etc which are designed to organise the data flow. In situations where there is a WAN this gets more complicated. If the WAN involves a shared technology (such as MPLS, DSL etc) then key resources in the network are being used by multiple data sources.
All these factors add up to the fact that the realistic throughput of the network is not the theoretical maximum, but exactly how close can you actually get? Understanding the real throughput of a network is key to knowing when the bandwidth is under-utilized and when you have reached the limit.
Other factors which come into play are; poor routing configurations (how do you know if you are using a hosted service?), MTU conflicts (which are common in encrypted networks), routers which are locked down too much and resources being utilised by a third party sharing a common internet router.
Point to point network performance can be measured by the standard RFC 2544 (with tools such as Fluke Networks EtherScope). However these are intrusive tests and can often miss fluctuations in performance. If you need to monitor effective performance all the time then a non intrusive test (such as Apparent PathView) becomes vital.
A good example here would be an 8Mbs DSL service versus an 8Mbs MPLS service versus an 8Mbs leased line. On paper they all promise 8 Mbs, however in practice the leased line gives the best performance, but why? The DSL service is certainly the cheapest so what can it offer and is it good enough?
Although many tools can measure the bandwidth of all these services, the effective performance of each requires specialist performance technology (such as Apparent PathView).
Return to Article
Last Updated: 08/12/2009