APM pays off

You can have better results today.

Naive visitors to “IT Ops” think it repeats the same message: a couple times a week, I provide tips on application performance management (APM) or comment on trends in datacenter operation. Many of the titles of individual postings would be just as timely three months in the past or future. At that level, it’s true the message rarely changes!

It doesn’t get old for me, though, because so many worthwhile APM solutions remain underapplied. The topics which appear here make a bottom-line difference to real-world applications; that keeps my enthusiasm up. Note these solutions are not about trying to catch the right fashion, or guess a market-shift that might happen; APM, done right, pays off demonstrably, measurably, and quickly.

A recent report from Joshua Bixby of Radware illustrates this. His team has gone to the trouble of timing performance of 400 top European retailers: “The median page took more than 7 seconds to load.” Similar results are available for North American operators, incidentally.

Bixby documents that we have abundant evidence that consumers respond to pages that load in about two seconds— maybe three seconds, or half a second, but seven seconds is well into the range that loses customers’ engagement. As much as we know anything about on-line operations, we’re sure that two- or three-second delays serve end users far, far better, than seven- or eight-second ones.

That’s not all we know, though: it’s equally certain that a few straightforward technical fixes, such as employment of a content-delivery network (CDN) and simple configurations like enablement of HTTP compression, often yield substantial, easily-measured gains. Turn on gzip; accelerate display of your Web pages; and increase the satisfaction of your Web visitors. As easy as that recipe sounds, it remains surprising how many organizations appear not to have put it, and similar ones, into play.

Why does APM have so many easy improvements? That’s a tougher question, and one that gets my attention more rarely. Here, too, though, at least a few pertinent measurements are available. A report titled “Top Challenges Performance Engineers Face in 2013”, from Shunra Software, finds that “61 percent of [survey] respondents do not design or develop for performance.

At a casual level, we all can understand that organizations need to ensure first that basic functional requirements are met. That leaves abundant “low-hanging fruit” for the DevOps engineer with an interest in improvement of performance. Most deployed applications are far from achieving their performance limits.