Why Application Performance Management Matters Now

Today's average business application typically runs in a virtual machine, with several branches that go off the application server and out to a completely different application for a particular function or piece of data.

In some cases, that branching operation may not even be on premises. As cloud computing takes hold, the scenario of going out of the network to obtain what your application needs will become more commonplace, leading to more headaches for the struggling art of application performance monitoring.

It's always been difficult to see what's going on inside the application. At the same time, doing so has become more crucial as applications become one of the main ways that customers interact with companies they wish to do business with.

The modern application performance management (APM) system requires an ability to see an application and its dependencies, compile statistics on normal operations, perform real-time analysis to detect anomalies, and do diagnostics to determine what can be done to correct what's gone wrong. No systems perform all functions perfectly, despite bringing some new strengths to the process.

One of the lesser-known APM systems, OpNet, in Bethesda, Md., has a product line that corresponds to several of these steps, including AppResponse Xpert for monitoring the end user experience, network and application; AppTransaction Xpert for "deep transaction performance analysis and prediction," according to its descriptor on the OpNet (now part of Riverbed) website; and AppMapper Xpert for mapping an application's parts and dependencies.

[InformationWeek's Art Wittmann argues APM suffers because vendors haven't kept up. See What's Killing APM?]

Covering the bases is one thing. Getting the end-user experience down cold, so that the IT administrator has an idea what the user is seeing as an application runs, remains a challenge. CA Technologies, with deep experience in dependent legacy systems, has also moved onto this ground with its own Internet monitoring from "77 points of presence," said Dave Cramer, CA's VP of product management.

Dummy transactions that mimic real ones can be launched from those 77 points of presence and response times monitored. If a customer is running an application in the cloud, CA's points of presence offer response times between that application and points of presence in 40 countries around the world. Cramer said CA is still building out its independent monitoring network and adds 6-7 more points per quarter.

The granddaddy of independent monitoring networks, however, is probably that of the largest pure-play application performance management vendor, Compuware. It springs from Compuware's Gomez acquisition and includes 150,000 points of presence in 168 countries. The computers at these points can launch dummy transactions and record response times, assembling a picture of cloud vendor, ISP and Internet performance that can tell a business much about what's going on with its application.

An ongoing problem, nevertheless, of seeing what the end user sees is caused by the frequent changes in browsers. Once dominated by Microsoft Explorer and Netscape Navigator, the field now includes Google Chrome and Mozilla Firefox, as well as Apple's Safari, open source Opera and several other entrants.

"Firefox and Chrome change multiple times a quarter. Application monitoring needs to recognize how browser displays differ. There's a cascading set of changes and drivers" that the application performance management system must keep up with, John Van Siclen, general manager of the APM business unit at Compuware. One of the advantages of implementing such a system is that it relieves the enterprise from tracking all the browser changes, he said. APM provides enough background intelligence "to survive this wave of complexity," at least in Compuware's case, said Van Siclen. Several young companies are bringing APM systems to address the problem, including AppDynamics, New Relic and Manage Engine. In addition, Microsoft, BMC, IBM, HP and Dell's Quest Software are already well-established players in the space. Whether newcomer or old hand, all parties agree that is necessary -- but difficult -- to get an end-to-end view of how an application is running, from browser window down to the legacy application code itself.

New Relic was founded by CEO Lew Cirne, who was also a founder of Wily Technologies, acquired by CA as the basis for its APM in 2006. Wily's Introscope provided insight into Java applications, and most APM vendors today still provide the best real-time insight into running Java applications. (CA also can monitor and report on .Net applications, Cramer said.)

Introscope and the APM systems that followed it started to collect massive amounts of data from server logs, then analyze and apply it to ongoing operations. A great deal is known about the running characteristics of Java, and its related languages capable of running in the Java Virtual Machine, such as JRuby. One challenge for each APM vendor is to ask how deep its system can go into Java application diagnostics and how broad is its reach into applications that may not be written in Java.

Dan Kuznetzky, principal of the Kuznetzky Group, in a column Monday described several other points that can be used as metrics for evaluating APM systems.

-- Determine whether an APM system can automatically learn about database engines and application frameworks as well as the application, servers, storage and networking. What about future applications?

-- A wild card in APM is whether the APM system can recognize what mobile devices are being brought to work and the applications that run with them.

-- Test the learning capability of the APM system. Can it collect data from your running applications and learn from it? Does it "just provide alerts and reporting," or can it "actually manage workload components?" he said.

The ability to see how a running application is performing is a more crucial task than ever, but it's gotten more complicated. Slowdowns inevitably occur. Not being able to see what's going on in the end user's browser may turn into a new Achilles heel of IT administration, especially when all components of the application are no longer on premises.

APM is starting to hinge on an ability to use large amounts of operations data in near real time; that is, as problems develop. It's still an imperfect process, with much work to be done, but application management now is much less of a passive process of reacting to what's gone wrong, and more one of active intervention.