Web Performance Challenge

Software Magazine, August, 2001 by Tim Grieser

Quality of the end-user experience takes precedence

THE WEB CHANGES EVERYTHING, including performance management. Now that users have direct contact with Web-based applications, performance management needs to go beyond optimizing infrastructure elements to address the overall quality of the end-user experience, especially end-to-end response time.

Web-based applications dramatically raise the importance of the end-user performance perspective and end-to-end response times. This means that Web performance management must address such areas as monitoring Web transaction response times, identifying exceptions to transaction performance objectives, and identifying infrastructure elements that are causing performance exceptions. The ultimate challenge to software vendors and service providers is to automate this process--and the corresponding processes of problem identification, problem resolution, and return to service-level compliance--to the maximum extent possible.

Performance management has long been an operational requirement needed by IT organizations to deliver good service. In classical mainframe environments, the focus of performance activity is on managing large centralized OS/390 architecture processors. Here, the emphasis is on managing the utilization of large processors to optimize performance and to deliver good response time for batch and interactive workloads. Indeed, some early approaches to mainframe capacity planning essentially amounted to determining when a processor or group of processors would reach 95% utilization, under growth scenarios, and thus require upgrades.

As systems and applications evolved from mainframes to distributed client/server and now to the Web, the performance management focus expanded from managing large central servers to tracking and optimizing a diverse set of interconnected infrastructure elements, including PC clients, networks, Web servers, application servers, databases, and storage devices. Performance management software vendors are now racing to cover larger portions of the total infrastructure--the applications and database servers behind the firewall, the Web servers, the network, the PG client, and so on.

Even in distributed architectures, much of the performance management activity still focuses on managing individual infrastructure elements. The idea, of course, is to optimize the performance of each element--client, network, server--and thereby deliver good performance for the applications using them. Considerable amounts of performance management software solutions have been developed and deployed for such element management.

From Elements to Experiences

However, the shift to distributed Web-based applications not only magnifies the number of infrastructure elements involved in delivering an application (and hence requires performance management software) but changes the perspective from which performance is viewed. With Web-based applications, the performance focus is moving from individual infrastructure element management to optimizing the experience of an application's end user--who may be a direct customer or buyer.

The quality of end-user experience--good response time or annoying delays--often determines whether a user completes a transaction successfully or becomes discouraged, abandons the transaction, or even goes to a competitive Web site. In the e-business world, system outages or poor response times can lead directly to lost revenue. Given this perspective, it is easy to see how the continued growth in e-business and Web-enabled applications creates increased need to manage response times from the end user's point of view. As a consequence, managing Web performance challenges software vendors on a number of fronts.

A rapidly growing number of software tools vendors and management service providers are addressing the need for operationally monitoring and managing Web site performance. Some of the more innovative efforts are in the area of performance monitoring and load testing of Web sites as viewed from the end-user perspective. A typical approach is to send test transactions--from one or more PCs at remote locations--over the Internet to a target Web site. Time stamps are recorded and used to calculate various performance metrics, such as home page load times and transaction response times. While there are now a substantial number of vendors with solutions in this area, perhaps the best-known early examples are Keynote Systems for remote performance monitoring and Mercury Interactive for remote load testing.

Simple tests, such as the standard "ping" or "traceroute" commands available on network-attached PCs to measure packet transit times can be used to determine whether a Web site (URL) is basically operational and accessible. Synthetic transactions (often scripted versions of typical Web site requests) can be used to measure more detailed performance information, such as end-to-end response times for specific transaction processing activities. The interesting point here is that performance monitoring is being conducted from "over the Web" to obtain measures from an end-user perspective, rather than from an infrastructure-element perspective.