September 25, 2000

Printer ready

Web-Site Monitoring Derails Problems

The key is knowing what to watch for and how to interpret the data

By Billie Shea

More on Web-site performance:

sidebar: To Optimize Performance, Look At The Last Half Mile InternetWeek: Anticipate Problems, Map Changes (9/4/00) InternetWeek: E-Commerce--Stay Afloat Without The Bloat (8/14/00) VARBusiness: Optimize Your Client's Site (8/7/00)

TechEncyclopedia:

Need a tech term defined?

Send Us Your Feedback

here's no arguing that performance matters for those doing business online. Web-site performance is among the top challenges facing online companies. Bill Gassman, a senior analyst at Gartner Group, calls Web-site performance management "a multifaceted challenge" and says that "delivering the proper performance for E-business Web applications requires a continuous collaborative effort to properly define and fund business requirements, to apply the proper application architecture and resources, and then to monitor both user perception and infrastructure statistics through performance metrics."

Performance monitoring works to gather those metrics--the information about how a Web site is servicing its customers, suppliers, distributors, and business partners. Monitoring can alert companies to site slowdowns and failures, help isolate and correct performance problems, uncover trends, tighten Web-site design, set realistic performance baselines, and aid in capacity planning. However, knowing how and what to monitor, what tools and vantage points to choose (and when), and how to correlate and analyze reams of performance data to quickly pinpoint and solve performance problems is a complex task for event-driven, highly distributed Web environments.

There are multiple dimensions to a good performance-monitoring strategy. Companies must work to make sense out of metrics gained from infrastructure component monitoring (tools and services measuring activity specific to the network, clients, or servers from within the company's IT department) and those gained from the user perspective (tools and services measuring response times and throughput from inside or outside the company's IT department).

From each perspective there are multiple tools and approaches. None of them has emerged as a standard, but many of them are being used together to provide a 360-degree view of Web-site performance. However, from a data correlation, analysis, and interpretation standpoint, there's still work to be done to introduce more automation into combining metrics from varying vantage points. For the most part, the performance data gathered from multiple sources is a puzzle that requires a skilled IT staff to solve. And it's the most critical part of performance monitoring, because correctly interpreting performance data and turning it into actionable tasks that deliver performance gains are the keys to ensuring that monitoring practices return business value.

Therefore, the most balanced strategy for Web-site performance monitoring today is a combination of both the technical infrastructure view and the user-perspective view.

Bigchalk.com Inc., an online learning-tools provider that supports the K-12 education and library markets, manages its Web-site performance on the infrastructure level and the application level, says Mike McCosker, project manager of infrastructure. "Our application-level monitoring is an extension of our testing process. We also do internal programming work to correlate metrics from the various performance monitoring tools that we currently use." Additionally, Bigchalk.com is looking at a third level of Web-site monitoring to characterize and analyze patterns relating to its site traffic and is looking to Accrue Software Inc. to help with specific details on Web-site usage monitoring.

The idea of monitoring from the user perspective has gained considerable attention and momentum because of the Web's distributed nature. Companies want to understand how users experience their Web sites. Recently, options for getting a pulse on a user's perception of Web-site performance have emerged from traditional testing-tool vendors. Companies such as Compuware, Empirix (formally RSW Software), Mercury Interactive, and Segue Software offer products and services that extend the value of automated testing with Web application monitoring. This is necessary because even though application testing before deployment does well to shake out major Web-site showstoppers, problems can still creep in from unanticipated places. For example, continuous database logging can sometimes degrade application performance in the live environment.

To extend testing with monitoring practices, tool vendors leverage transaction simulation--the collection of round-trip response-time metrics for key business transactions conducted by a synthetic user. Transactions can be simulated from either inside or outside a company's IT department, and, in some cases, vendors provide the ability for both. Just like the large number of users that are simulated for large-scale load tests and fired against the system to test performance and scalability, those same simulated users can be fired individually at regular intervals to measure response times while a live site is being stressed with real users. With this approach, businesses that want to know what their users may be experiencing become their own users--in a simulated fashion--to constantly take sample measurements as real Web-site traffic volumes fluctuate.

For those who choose to approach Web application performance monitoring by extending their testing efforts, test scripts can be reused to simulate multiple transaction types (business processes). It's important to prioritize and measure multiple Web-site transaction types (logon, browse and search, purchase, account query, file download, etc.) on a continuous basis. Companies can choose to have transactions fired off as often as they like to capture and compare response-time measurements, as well as verify correct page responses or transaction accuracy.

Further, the performance metrics gained before deployment can serve as baselines, which the simulated monitoring transactions should be continuously measured against. This method can give a good understanding about whether an application is meeting its performance objectives. If problems exist, alerts can be triggered. This approach serves well as a first line of defense.

At a high level, this approach can alert businesses to a performance issue and can determine whether the issue is within a company's firewall or outside of its control. If the root cause is in the infrastructure, this approach typically won't provide the level of detail required to uncover the cause--that's where specific infrastructure point solutions and the IT team's expertise will come in.

Performance trends uncovered with continuous application monitoring or testing can be used to improve future application performance and scalability tests by adding a higher degree of realism. Knowing how users may be experiencing your application provides valuable information for modeling realistic user loads for performance testing. It can also give an understanding about how specific Web-site design changes could enhance performance, and it constantly verifies that tested applications deliver adequate and consistent performance according to expectations.

Monitoring in this way can be an early indicator of when system expansion may be necessary to accommodate business growth in terms of higher volumes of daily traffic. Testing then becomes a more cyclical and evolutionary process, spawning continuous quality improvement.

"Because there is such a push for reaction to the market and time-to-market and because Web applications change so frequently, the reliability tests that would have traditionally been part of the pre-deployment testing phase are now being spread around the development cycle and carried into production," says Rob Cohen, VP of product marketing for Segue Software Inc. Based on what Segue was seeing in the market, the company recently unveiled consulting services for Web-site monitoring--eCMonitor VPM (virtual private monitoring) and eCMonitor CVM (continuous verification monitoring)--aimed at helping business-to-business and online exchanges monitor performance and transaction accuracy across their networks of partners, customers, and employees.

continue on to page 2

Illustration by Marlena Zuber
Photo of Cohen by Brian Smith

Back to This Week's Issue
Send Us Your Feedback
Top of the Page

---