There’s so much to like in Michael Biddick’s “The Next Wave In APM” (TNWIA) that it resists easy abbreviation. You’re best off to read this InformationWeek report for yourself, and savor the many high points, including sections on “Complexity Leads To Frustration”, “APM is still useful, but success means being smart, being selective and looking beyond legacy tool providers”, and the challenge of “the migration to virtualization and cloud delivery models”.
TNWIA amplifies many of the points “Application Monitor” has made in the last couple of months about the position of APM, prospects for the future, and the role(s) of metrics. Biddick particularly emphasizes the fundamental split in the application performance management (APM) marketplace: some applications simply aren’t important enough to merit the cost in complexity and management overhead of APM, while others are so crucial that APM is indispensable. The returns to APM are highly non-linear, and we need to keep our attention always on on the areas where APM pays off–because, when it does, it pays off big.
Note also that “important” in the previous paragraph has a specific engineering sense that decision-makers sometimes don’t grasp. E-mail, for example, might be crucial to an organization: the organization goes to great lengths to ensure the availability and reliability of e-mail. E-mail’s performance, though, might matter little, in the sense that it doesn’t vary in a matter which affects business operations: whether an individual item arrives within fifteen hundredths of a second, or fifteen minutes, is of little consequence. APM is for the applications whose performance matters, and, for those, the penalties of delay can be severe. Although some IT shops don’t get it yet, these rewards and penalties remain even if all consumers are “internal”. As Biddick puts it, “… organizations need to start acting like true service providers or risk being outsourced by unhappy customers.”
Immediately after this, Biddick mandates that “we need to focus on simple yet telling metrics …” Here again, even common words are subject to mis-interpretation. “Simple” might not be easy from an engineering standpoint; the intent is more to choose metrics that are understood transparently by all actors, and that contribute to a compelling dashboard.
Other strengths of TNWIA include its explanation of reactive vs. proactive APM, the importance of application dependency mapping, and the likelihood of turnover among vendors amid the rising importance of open-source and lightweight tools. Biddick particularly elaborates the challenges that virtualization and the cloud present to APM, and how far existing products have to go, even in their licensing, to handle virtualization intelligently. He’s negative on “[c]loud-based management tools [that] will never really be an option for APM given the amount of traffic that needs to traverse the Internet …” This particular claim leaves me unconvinced; while he might be right, I also can imagine clever techniques that eventually make APM-as-a-Service a winner. More surprising to me is that this particular report doesn’t mention software-defined networking (SDN) at all as a challenge for APM. To me, SDN is as likely to stress existing solutions as much as host virtualization has.
Also hard to judge: “[t]he convergence of performance management tools that are focused on development environments and those that are focused on operational environments …” Biddick finds the convergence “likely”. I know it’s desirable, and perhaps even overdue, but uncertain whether the vendors prominent in today’s market will deliver. In part, my uncertainty comes from my skepticism about the survey approach which backs up InformationWeek reports; I simply find it hard to draw strong conclusions from this kind of poll.