With good reason, key sessions at events such as this year's UK Computer Measurement Group conference (see www.ukcmg.org.uk )have been concerned with the issue of trending in Performance Management. E-business is the driver for all this concern. The belief is that due to some of the issues involved in managing e-commerce systems, trend analysis will offer vital insights for system and performance managers.

The concept of the 'transaction', the return of an old mainframe favorite concept, is often touted as the one reason for this. The belief, hopefully justified, is that the 'e' transaction will relate more closely to a business transaction, i.e. an enquiry or an order. Certainly it is valid to think of e-commerce applications as the latest embodiment of high volume transaction processing.

Evidence so far suggests that e-commerce applications will follow very regular patterns when viewed over time. A parallel is the clearly visible 'follow the sun' daily processing patterns of global financial applications. Despite the huge variety of different activities being carried out, the pattern of activity on the Internet in the UK throughout 1999 showed a more reliable and predictable pattern than is the norm for traditional in-house applications. A paper copy of the Independent article from last November supporting this statement is available from Metron by post. Please e-mail your postal address to techinfo@metron.co.uk.

Trends will be marked by significant workload spikes and transfer size variations though, and the manager will need knowledge and software tools to understand these. Workload spikes are expected to vary by as much as 8-10 times the average, more than one would expect for traditional applications. Traffic patterns will be characterized by 'power tailed distributions'. This means the traffic will consist of a very large number of relatively small (c. 1KB) packets and a small number of large (c. 1MB+) packets.

Understanding these workload trends will be vital for the performance manager in the process of performance analysis and workload prediction. If trends can be accurately identified and extrapolated, the resources required to support workload demand in the future can be quickly and accurately assessed. Good trend analysis allows the system manager to get ahead of the game, rather than be chasing it fixing crises.