Blueprint For Application Performance ManagementBlueprint For Application Performance Management

BUILDING APM
The APM architecture is built on three elements that enable testing and incident investigation capabilities: data collectors, analysis engines, and reporting stations. These elements come together to build a set of tools that proactively monitor systems and resolve application problems. In some cases, problems are diagnosed through active synthetic transaction monitors, while others may require passive agent or agentless monitoring.

Synthetic transaction monitors measure application performance by simulating user activity using predefined transactions. They can identify many user-perceived performance problems, but often can't determine where the actual problem is occurring. What's more, they require unique programming for each application monitored. Perhaps their most important use is for reporting user experience data, which can be trendable over long periods and through application revisions. Such data can be extremely useful for reporting on IT's service-level agreements.

Alternatively, or in addition to synthetic transaction monitors, IT can capture application performance data passively by deploying software agents and hardware probes. While these provide a more detailed picture of the underlying applica- tion operation, they also can incur significant deployment and installation costs, and take more day-to-day attention. Such systems are likely to observe and record events that actually cause undesired application performance, but finding those events and correlating them back to an observed performance issue is an evolving science.

Hardware probes attach at key network junctures, such as Internet access points or via switch monitoring ports, and are normally passive. They also connect to core switches and collect NetFlow statistics to gain a more complete view of the IP infrastructure. As such, these probes can gather a lot of data. To prevent the that data from inundating the network--particularly WAN links--analysis engines must be deployed throughout the infrastructure. These systems aggregate and process the data from the various probes and, depending on the size of the organization, consolidate data from a number of sites.

(click image for larger view)

Finally, a monitoring station, or management console, enables staff to query these various components from a single location. Numerous functions and technologies are made available within this context to the IT manager. Most important is the ability to analyze and correlate results from many locations such as the user's desktop, the network, and the data center. The monitoring station should be capable of problem resolution functions and service-level monitoring and reporting.

An APM architecture should assess the technology's performance against the actual user experience. Work being done by the Apdex Group (www.apdex.org) aims to standardize these measures. The group, spearheaded by Sevcik, seeks to provide a numerical measure of user satisfaction with enterprise applications. The organization aims to create specifications that calculate one number from many measurements on a uniform scale of 0 (no users satisfied) to 1 (all users satisfied) that can be applied to any set of user perception measurements.

At the same time, within the data center, metrics are needed to capture transaction performance as well as the performance of data center components. Across the network, traffic analysis using standards such as NetFlow, and, more granularly, classical packet analysis, lets engineers analyze the performance of the corporate network. Coupled with route analytics, which provides an understanding of routing's impact on application performance, IT managers gain a complete picture of application dependencies.

Ultimately, the goal is to gain a holistic view that accounts for the unique characteristics of each device and system required by the application. With a coherent view of the application's end-to-end performance, managers can better understand the implications of infrastructure changes on the application, support application life-cycle planning, and ultimately improve the ability to deliver what matters most: a satisfied end user.

Photograph by Jupiter Images