Taking performance management to the edge

Nowhere are details more devilish than in the realm of application-performance management, where the littlest of things — say, a subtle increase in the number of program errors or a gradual decrease in server responsiveness — can be harbingers of doom. And with all eyes fixed primarily on the datacenter, it’s easy to lose sight of that critical final variable in the application-performance equation: the end-user desktop.

Enter Reflectent Software’s EdgeSight, a Windows-based application monitoring and performance management solution that does what traditional management megaframeworks cannot — it identifies, diagnoses, and reports application performance and availability issues from a client perspective.

EdgeSight accomplishes this through a combination of detailed process analysis and historical trending, including dynamic baseline comparison (of network response times and application-usage patterns) and centralized alert tracking (for such problems as application hangs, missing DLLs or components, and program crashes). With EdgeSight, IT shops can find out not only why a fault condition exists but also how the current environment differs from what existed an hour, a day, or even a week earlier. User-definable reporting thresholds and event filters help keep extraneous data in check, allowing the truly critical items to rise to the surface.

Dizzying depths

The heart of Reflectent’s solution is the EdgeSight Console. Here you can view various summary reports — grouped by device, application, or event type — or you can drill down into specific areas for more detailed analysis. For example, an increase in alerts for a specific application might prompt you to further investigate the faulting process. Digging deeper, you discover that a specific DLL, which the application attempts to load into its address space, is causing an error due to improper registration. Additional details about the DLL vendor and version then make it easy to isolate the failure and resolve the issue.

Similar multilevel trending and analysis paths are available for network performance — proactive ping — and application-usage patterns. Coupled with myriad additional report views, such as a What’s New view for identifying recent changes to a given category, this multilayered design helps to further refine the flow of data. In fact, if there’s a flaw in the EdgeSight solution it’s the sheer number of report types and viewing options. The user is presented with a dizzying array of layered windows and expandable tables, and it’s not always clear how to “get there from here.”

For example, when viewing the top-level summary by Device, there’s no clear way to drill down to the specific report for CPU Utilization — itself a subcategory of the top-level System category. Assuming you can tackle the console’s somewhat intimidating GUI, you’ll begin to appreciate the level of correlation between application and system metrics and how the various layered views tie together these elements to paint a cohesive picture of client performance and availability.

I installed Version 3.1 of EdgeSight and proceeded to collect data from three Windows-based client workstations over several days. Installation of the server components was straightforward. Using Reflectent’s Web-based installer, I was able to download and configure the various EdgeSight pieces. The installer itself was intelligent enough to identify any missing prerequisites — in my case, the Microsoft Message Queue Service — and to offer various options for integrating with my existing infrastructure — for example, how to access the Microsoft SQL Server data store.

Agent awareness

After the EdgeSight server was in place, I was able to configure the client agents. The EdgeSight agent maintains a one-week, client-side store of near-real-time data — with a 3-second sample period — and uploads daily summary packets to the EdgeSight server via HTTPS — through direct connection only; proxy servers aren’t supported in this version. System administrators who want to access the more granular, client-side data can do so via an ActiveX control that runs from within the EdgeSight Console. This is a useful option for performing detailed analysis of client metrics.

During testing I simulated a variety of failure modes, including sudden spikes in client CPU utilization and various application failures, such as missing DLLs, crashes, and hangs. The EdgeSight agent faithfully identified and recorded each occurrence, uploading the relevant contextual data — including historical trending values for comparison — and allowing me to dig deeper into the events to find their root causes. It also recorded some useful security-related events, including USB-device activity — for example, the addition or removal of portable media, a major problem for environments that handle sensitive data.

Pricing for EdgeSight 3.1 is typical for a comprehensive management solution. Deploying the EdgeSight server components will run you $50,000, while each instrumented client device requires a $50 seat license. The company also produces various connectors for integrating with larger management frameworks. Finally, Reflectent recommends allowing one of its engineers to install and configure the solution, so budget for some consulting when figuring your initial cost of deployment.

Overall, EdgeSight delivers on its promises. IT shops gain significant insight into the behavior of their client run times, a need that the mainstream management community has largely ignored. Things I’d like to see in future versions include a more task-centric navigation model — for example, pulling together various report elements to create a high-level ROI or lifecycle study — and the ability to operate over proxy servers. Reflectent is promising to address the former issue in its next major release, EdgeSight 4.0, which will also feature a revamped menu layout and integration with Microsoft Reporting Services.

In the meantime, EdgeSight 3.1 does an admirable job filling the void between low-level, OS monitoring tools such as Windows’ Performance Monitor and the more all-encompassing but less detailed systems management frameworks.