In the UNIX world, performance is often one of the big decision factors, especially for scale up workloads. Over the years, I've taken part in a lot of performance discussions, both with customers and with the engineers that design systems and write software. While reliability, availability, serviceability, virtualization, management, and more are all important factors, performance influences everything from server sizing, hardware, operating system, and software licensing costs to ongoing power, cooling, floor space, and support costs.

However, analyzing performance, especially between vendors and different types of servers and applications, can be challenging. For instance, the very well known TCP-C benchmark is a great way to size transaction processing if you have no latency requirements and want a benchmark that is based on character-based terminals. This is a starting point, but let's face it, it probably doesn't reflect the workload you are actually running (for instance, do you have any queries?)..

On the other hand, vendors and customers don't have the ability to run realistic benchmarks for every single workload and publish them. So, how do you get an idea about what the real-world performance will be for a new server or application?

This will be the topic of discussion on an upcoming webcast, based on the 2009 UNIX servey results from Gabriel Consulting Group. If you are interested, more information on the web cast on November 17th is available at http://www.cio.com/webcast/502060/?source=ciocli.