Web Load Stress Testing: Are Cloud Services Reliable?
Someone must stress-test the cloud if we want to take it seriously, as a real IT resource for the real business. On the last week some Australian researchers took up the challenge, and the stress-tests showed that the infrastructure of on-demand services offered by Amazon, Google, and Microsoft suffers from regular performance and availability issues. Researchers have created a web load stress test that simulated 2200 simultaneous user connections to applications hosted on Amazon EC2, Google AppEngine platform and Microsoft Azure Cloud computing.
As always with these types of tests, there is some good news and bad news:
The good news is that the testing did confirm that the cloud computing platform could scale as needed and responded to the dynamic growth of demand load. In fact, when the demand increased, dynamic cloud computing, provided that the additional capacity needed to meet the demand.
The bad news is that performance is very different. Indeed, according to researchers, the response time during the test, differed by 20 times, depending on time of day test occurred.
Testing also showed that various computing platforms clouds were individually suited for certain types of applications. For example, AppEngine to Google, seems to work best in simple applications or tasks that take less than 30 seconds. Google tools for monitoring were more oriented toward the developer and not such a great fit for business users.
The results of testing are encouraging, considering that the great concern about cloud computing is the ability to dynamically scale, which is one of the main reasons to use computing cloud, in the first place. The possibility of increasing conversion on demand, without having to go out and wave to purchase equipment and software upgrades, indeed, why we look at cloud computing.
On the other hand, performance problems are daunting - but not surprising. Anyone who uses a Web-aware that remote sites have huge differences in response times, and in many cases there are any number of links in the chain beyond themselves remote servers, such as network saturation that can cause delays. Nevertheless, the researchers said that they took this into account in its testing methodology.