APM gets smart with ProactiveNet’s latest

Intelligent software is the Holy Grail of application development — especially in the APM (application performance management) arena, where system administrators are perpetually on the lookout for more efficient ways to sift through mountains of metrics and alert data.

The better a given management solution is at filtering out noncritical events — while still flagging potential trouble spots — the more useful it is to IT as a whole. That’s why I found ProactiveNet’s pitch so remarkable: It claims ProactiveNet 6.0 can intelligently filter the systems management event stream so that only items worthy of attention rise to the top — a capability dubbed Intelligent Thresholds.

ProactiveNet accomplishes this by establishing a historical baseline of normal activity for a given entity — such as transactions per second or CPU utilization on a database server — and then using this as the filter through which ongoing operational data can be analyzed. It’s an intriguing approach, one that seems to hold tremendous promise for organizations struggling to wade through their management data.

Intelligence Vs. Control

ProactiveNet establishes a dynamic baseline by updating data points over time to reflect gradual fluctuations in application and system usage patterns. This approach helps eliminate the definition and tuning of custom metrics thresholds, which is a tedious and tricky task: If you set threshold values too high you risk missing critical events, but if you set them too low you will certainly be inundated with irrelevant data.

ProactiveNet does away with the threshold guesswork, but it comes at a price in terms of absolute control over the event filtering process. This latter point is key: Regardless of the potential for resource savings, many IT shops will be reluctant to surrender so much control to an automated process, no matter how sophisticated the analysis component.

This is especially true when you factor in other aspects of the ProactiveNet framework such as its automated “corrective actions.” The idea of having a systems management solution take action based on a soft threshold that is dynamically generated — as opposed to a predefined hard value — is enough to send many old-school techies reaching for the antacid jar.

The folks at ProactiveNet openly acknowledge that customer skepticism is one of the company’s biggest challenges, and they’ve become experts at the proof-of-concept deployment. But other potential pitfalls are endemic to the dynamic nature of the ProactiveNet approach.

As I mentioned above, ProactiveNet uses baseline data collected from the environment when it filters event and metrics data for possible anomalies. Such an approach assumes that you’re starting with a healthy system. Introducing a solution such as ProactiveNet to a failing environment could potentially short-circuit the analysis process, because the baseline data will already be out of whack.

To address this scenario, ProactiveNet 6.0 provides numerous traditional threshold-analysis methods for use as possible fallback measures. Given the potential for inadvertent “spoofing” of the analysis data, however, it seems clear that this is a solution suited more for strategic, long-term application maintenance than for tactical problem resolution.

Raves for Quick Console

I tested ProactiveNet 6.0 using a Sun Solaris system provided by the company as the back end and a custom, three-tier, Windows-based (ASP) client/server solution as the instrumentation target. Installation of the various agents and services went smoothly, and I was able to quickly build a baseline data set against which I could compare various anomaly scenarios.

ProactiveNet’s user reporting interface — dubbed Quick Console — is entirely Web browser-based, which helps hide the complexity of its Solaris underpinnings. Using Internet Explorer, I ran a variety of informative reports, including an analysis of SLA compliance by geographic site location.

Numerous event filtering and drill-down analysis options make investigating a specific anomaly a straightforward process. It quickly caught significant deviations from baseline for CPU and memory utilization on Web and SQL servers, flagging them for investigation.

I especially liked how ProactiveNet was capable of breaking out SLA results by time period, with data on daily, weekly, monthly, and various “to date” variations. Paired with ProactiveNet’s intelligent event filtering, which looks for trends across multiple sample periods to weed out spikes and routine utilization surges, the Quick Console left me feeling empowered and entirely in control of my simulated enterprise.

Pricing is typical for this category. A simple, bare-bones three-system/three-tier deployment with Web server, database server, and application server will cost you less than $15,000. But the pricing scales with the number of monitored devices and can quickly climb into the six-figure range for a large organization with more than 100 monitored entities.

Overall, ProactiveNet is a well-crafted solution that promises to take some of the tedium out of APM. The Quick Console interface is excellent, and it does a good job of serving up the most important information from a pile of performance data. The only downside is the requirement for Sun Solaris running on a Sparc box for ProactiveNet 6.0’s back-end operations. This more than anything else has the potential to hinder broad-based adoption.

Fortunately, the company is working on a Windows-based version and hopes to have some non-Solaris/Sparc options available next year. In the meantime, customers unfamiliar with Solaris-on-Sparc will need to take advantage of ProactiveNet’s consulting services for initial setup and deployment.