Elementary Performance Modeling – Lesson 1

Elementary Performance Modeling 101 – Part 1 of 5

Lesson 1 – Elementary Performance Modeling


  • 2

    Handling parallelism



    At one point in this lesson Dr Rajesh says that one way you can reduce demand is by increasing resource; in this case he talked about doubling the number of database CPUs. Am I right in assuming that you would have to have a program architecture that could benefit, i.e. enough threads to occupy the additional CPUs?

    Is there a systematic way of including parallelism factors (e.g. multiple CPUs) into the demand calculation?

    Thanks and regards…Paul

    • twarren

      Yes, you are right in assuming that the underlying application/database should be capable of leveraging the additional compute capacity thrown at it. With enterprise Databases like Oracle, DB2 you shouldn’t have much of an issue. Obviously, expecting linear growth in scalability is something one should be wary of.

      I’ll forward the request to Dr. Rajesh and let him chip in.


    • twarren

      Dr. Rajesh Says – For this particular case, Avg Svc Demand per Resource = Visit Count x Avg Svc Time. Having more CPUs means less visit count per CPU and hence lower service demand.

      It is fine for any given service to run single threaded as long as you have enough workload to run multiple services – eg standard 3 tier OLTP architectures.