This paper discusses software performance in a very practical light and addresses the problem of optimizing performance in a way that developers can understand and apply across a very wide variety of applications and solutions. Its goal is to give the average software developer and IT professional a working knowledge of the process, tools, and best-known methods required for diagnosing, root causing and addressing performance problems in modern software applications.

Start here: beginning to quantify performance
Often, the most difficult task of all is simply identifying the need for performance engineering in a given product. Without a basis for comparison, one could argue whether or not the need even exists. If, for no other reason, consider the end user. Whether that person is an IT professional managing hundreds of servers or a PC gaming enthusiast, gaining more performance from software results in cost savings for everyone. Cost savings means a better overall user experience and simpler solutions that yield greater ease of management. For these reasons and many others, performance should be a primary consideration, rather than an afterthought.

In order for software performance to become practical, it must first be quantifiable. What exactly is required? First, one must identify a metric for measuring an application's performance. This may seem to be an easy task, but it is more often than not quite complicated. Some software applications simply do not lend themselves to measurement. Moreover, even if a metric seems readily apparent or obvious, how does one know it is the right thing to measure? The only way to answer this question is to ask more questions:

• What does the application do? (i.e., calculations, analytics, modeling, graphics rendering)
• What system components are stressed during runtime? (i.e., CPU, memory, graphics, disks, network)
• With what metric might the end-user be most concerned? (i.e., startup time, request-processing time, user experience)
• Are there several components or tiers that the performance engineer should consider? (i.e ., Web, database, application, network services)
• What comprises a ' representative ' end-to-end solution? (i.e., interaction between multiple systems or nodes)