It’s 10 p.m. — do you know what your APIs are doing?

Do you know what cloud services and APIs your systems are relying on? And do you know how to tell how reliable they are, and whether your vendor’s service level agreement (SLA) covers the performance you need?

1 2 Page 2
Page 2 of 2

You also need to look at how you treat API calls once the response reaches your network. “People assume that APIs are just Web pages; they’re not,” O’Neill explains. “Just because it’s an HTTP call doesn’t mean it works like or can be treated like a Web call.”

U.K. energy provider First Utility was using a range of testing and monitoring tools to manage the APIs it uses to build the iOS and Android apps its customers use to monitor their energy usage in real time with smart meters, as well as customer account services on its website. But when First Utility tried the APImetrics service, it quickly discovered that its internal architecture didn’t treat API calls the same way as Web traffic and was actually throttling them, affecting the experience users were getting. Another APImetrics customer found that while Web traffic went through its enterprise service bus exactly as expected, API calls going through the same bus experienced significant delays.

In another case, the problems that seemed to be caused by an API were down to network logging software that was running out of disk space once a week: Every Tuesday afternoon while it was in maintenance mode deleting logs to make space, all the API calls it was logging were stuck in a task queue. That would trigger lots of complaints from customers, but when the engineers checked the service the next morning, the API and the system that called it would both be working normally again.

[Related: First look: Microsoft’s API mashup tool for the rest of us]

Network security, like deep packet inspection, might also affect how your network handles API calls and responses.

“It’s about taking an outside-in approach,” says O’Neill.” Until you focus on the end-to-end experience, you will not know how APIs are functioning for the apps you build, for your own users or for your customers.”

Is it slow or is it unusually slow?

Not all clouds are equal. According to measurements that APImetrics has made over the past few years, if you’re using DNS routing inside Azure, you’ll see worse performance. “What takes 10 milliseconds in an AWS cluster takes 60 milliseconds inside an Azure one,” O’Neill says. “When you switch to using IP addresses though, the Azure speed is the same as the AWS speed.”

Developers using APIs need to consider these issues and start choosing between different APIs and cloud services based on responsiveness as well as functionality and cost. And ops will need to monitor API responsiveness over time. That’s not necessarily happening today, O’Neill says, “because it's hard to measure, and you need to think about where you're measuring from, as well as what and how. It's not getting measured, and it’s getting brushed under the carpet.

outlier 2b Microsoft

“Outliers are items that would effectively have fallen so far outside the statistical parameters that there’s no point pretending they actually worked,” says APImetrics CEO David O’Neil.

For that to change, and to make API performance management more common, business leaders will have to start caring about those questions. “It’s all about taking issue-detection and performance monitoring out of DevOps and putting it in the hands of business managers and owners where it really belongs,” O’Neill argues. “Revenue-generating, customer-facing teams will be the ones who will be responsible for this stuff, and they're going to call the shots.”

To serve that audience, APImetrics is building metrics that are designed to be more comprehensible for business leaders who aren’t networking experts. One approach is a number indicating the reliability of an API. O’Neill compares it to a credit score. “One API might be slow but consistently slow. You need to worry about the variation from what you expect rather than the speed,” he explains. “If 5 percent of the time it’s too slow to use, then that's a problem.”

Another useful exercise is to look at outliers: How many calls have been exceptionally poor?

“What you need to know is not whether the API is up, but whether it's functioning well enough to serve your customers. That sounds like the same thing, but they’re subtly different,” O’Neill cautions. “You need to be able to see what's statistically normal and when you're not getting that. You might not like what’s statistically normal for your API, but that's a different problem.”

Copyright © 2016 IDG Communications, Inc.

1 2 Page 2
Page 2 of 2
Download CIO's Roadmap Report: 5G in the Enterprise