Performance engineering

Failure caused by poor performance renders a system just as useless as failure caused by functional errors, and can be even more expensive to correct. For many years, performance in software systems has been achieved through a ‘fix-it-later’ approach, using optimisation and tuning techniques. Recent evidence shows that, with the increased use of highly complex, multi-layered, client/server, distributed architectures, these techniques cannot be relied on to deliver even the level of performance required during early operation when workloads are small, let alone over extended periods. This paper describes some of the ways in which performance can be ‘engineered’ into systems — examples of how these techniques have been used in recent BT projects are included.