Cloud computing utilizes computing resources, network bandwidth, storage, applications, and services available in the Internet "cloud" to deliver scalable Web functionalities to end users anywhere in the world. Drawing on the cloud for computing resources is similar to tapping into the electric grid for electricity - cost is incurred only as resources or computing cycles are consumed.

Application owners can theoretically take advantage of the highly scalable infrastructure available from vendors like Amazon, Google, and IBM and services available from application vendors like Google, Microsoft, and Salesforce to deliver application functionalities without incurring capital expenditure, the headache and expense of operating a data center, or the cost of developing common application functions like billing, shopping carts, and CRM. In effect, application owners can focus on delivering their unique value and rich user experience and leave the mundane development and management tasks to the domain experts.

But is it so simple?

Not really.

Cloud computing depends on a loosely coupled amalgamation of hundreds of hardware and software modules or services from multiple third-party vendors. As a result, IT has no direct control over this infrastructure. Nevertheless, IT is still responsible for application availability and response time.

As the old adage goes, you manage what you can measure. So it follows that IT must have tools that accurately measure application performance from the perspective of the end user to ensure application response times meet the requirements of the end user. That means that for applications delivered via the cloud, the point where you measure performance will dramatically alter the data. Understanding where application performance is measured, basically determines what performance data are available and that in turn impacts the validity of the measurement.

Where and What
What is an application performance problem?

Simplistically, you have a problem when an end user isn't getting the transaction response time he or she expects or needs to complete the job at hand. For an e-commerce site, that might be the time required to search for and display the image of the ideal little black dress. With consumers being more demanding and with increasingly intense competition among e-commerce sites, many studies indicate that users who wait more than about four seconds tend to click away to a competitor's site.

In the recent past when all hardware and software components existed in a single data center all under IT's direct control, it was possible to assume a strong correlation between the performance of network and servers and user experience. This made the where and what question simple: Where? In the datacenter. What? Server performance.

Cloud computing fundamentally changes this equation. While it's nice to measure CPU and memory consumption for the Web server that delivers the image of the little black dress, these metrics have little bearing on the time the customer has to wait before the page components wind through the cloud and load in the browser. For that matter, these metrics won't tell you if the image was ever delivered. Nor will you know if any of the other objects on that page served properly. These metrics simply can't comprehend the hundreds of hardware and software components and services involved in serving up the final page. Any of these components or services might degrade page load times or worse, leave gapping holes in functionality. Imagine the impact on sales when the "Buy Now" button or the link to the shopping cart never makes it to the screen.