Most performance monitoring systems focusing on E-business infrastructure measure delays in the web server, app server and network infrastructure. The back-end database, however, is often the bottleneck in transaction heavy web environments. Typical tools treat the database as a "black box," offering no visibility to the internal database resources and their delay impact on user response. A new class of tools measures wait-time for hundreds of specific delay points inside the database, and quantifies exactly what response time boost you can achieve by resolving the bottleneck. By looking inside the database and watching the wait-time created by specific transactions, you can make major performance gains without adding new hardware capacity.
Watching your Wait
If you needed to improve the speed of your morning commute, would you start by measuring the number of times your wheels turned on the drive in to work, then compare with your tachometer and temperature gauge? No, of course your best measurement is on the clock – how long did it take to get there, and where did you spend the most time? At the stoplight? On the freeway?
Surprisingly, many IT professionals and management still make decisions by watching event counters or “dashboard gauges” the same way, and use this information to attempt performance optimization or to validate capital investment projects.
Modern performance management methods have gone past counting specific transactions or I/O operations and now focus on wait-time as the most significant determinant of performance. End-users primarily care about one thing: how fast the system returns their transaction. Forward thinking IT shops now also primarily focus their tuning efforts based on impacting the perceived response time for the system. SLAs to external customers or internal business units are more frequently being impacted by the time it takes to deliver service.
An effective performance management system must utilize wait-time as the primary unit of evaluation, and allow performance optimization to minimize the wait-time experienced by end-users.
Where do you Measure?
Once you know that wait-time is the key statistic, where do you measure it? A typical end-to-end e-business infrastructure includes network components, web servers, application servers, and back-end databases. When monitoring the wait-time performance of web transactions, most of the attention is placed on the front end, possibly because it is easier to look there. The systems are simpler and easier to instrument for measuring inbound and outbound traffic, and the internal processes are far less complex than in a database that must maintain fault tolerant records of every transaction. Why look for problems where they are hard to find?
-1-