Capacity alone won’t assure good cloud performance

analysis
Feb 6, 20182 mins

Will your cloud-based workload support 100,000 concurrent users? The only way to find out is through performance testing

analyzing performance of wireless mobile connectivity data statistics
Credit: Thinkstock

Many people believe that workloads in the cloud always perform better because public clouds have access to an almost unlimited amount of resources. Although you can provision the resources you need—and even use serverless computing so the allocation of resources is done for you—the fact is that having the right amount of resources is only half the battle.

To get good cloud performance means you have to be proactive in testing for performance, not be reactive and wait for an issue to arrive in production. After all, performance depends on much more than raw capacity.

I strongly encourage testing. If you’re using devops to build and deploy your cloud application workloads, your testing for security, stability, and so on are typically done withcontinuous testing tools as part of the devops process.

But what about performance testing?

Truth be told, performance testing is often an afterthought that typically comes up only when there is a performance problem that the users see and report. Moreover, performance usually becomes an issue when the user loads surpass a certain level, which can be anywhere from 5,000 to 100,0000 concurrent sessions, depending on the application. So you discover a problem only when you’re got high usage. At which point you can’t escape the blame.

An emerging best practice is to build in performance testing into your devops or cloud migration process. This means adding performance tests to the testing mix and look at how the application workload and connected database deals with loads well beyond what you would expect.      

This means looking for a performance testing tool that is compatible with your application, the other devops tools you have, and the target cloud platform where the application is to be deployed. Of course, a “cool tool” itself is not the complete answer; you need testing engineers to design the right set of testing processes in the first place.      

Ironically, although devops itself ( as both a process and tool set) is all about being proactive in terms of testing, most devops processes that I’ve seen don’t do much performance testing, if any at all.     

Withouth that testing, you can’t answer the question “When will my cloud workload hit the performance wall?” Instead, your users find out for you, and you may discover it’s time to look for a new job.         

David Linthicum

David S. Linthicum is an internationally recognized industry expert and thought leader. Dave has authored 13 books on computing, the latest of which is An Insider’s Guide to Cloud Computing. Dave’s industry experience includes tenures as CTO and CEO of several successful software companies, and upper-level management positions in Fortune 100 companies. He keynotes leading technology conferences on cloud computing, SOA, enterprise application integration, and enterprise architecture. Dave writes the Cloud Insider blog for InfoWorld. His views are his own.

More from this author