4.6 Performance Models

Performance modeling is a key technique to understanding problems in IT systems. Because it is difficult to estimate performance, IT systems must be designed with service levels in mind. In other words, a designer of an IT service must know the limits of the system a priori. For instance, a designer must know the maximum number of transactions per second the system is capable of processing (i.e., an upper bound on throughput) or the minimum response time that can be achieved by a transaction processing system (i.e., a lower bound on response time).

Analytic performance models capture fundamental aspects of a computer system and relate them to each other by mathematical formulas and/or computational algorithms. Basically, analytic performance models require input information such as workload intensity (e.g., arrival rate, number of clients, and think time) and the service demand placed by the basic component of the workload on each resource of the system. Several queuing network-based algorithms for solving open and closed models with multiple classes are provided in Part II of this book and are implemented in MS Excel workbooks. The techniques include exact and approximate solutions. Many times, a relative or approximate performance estimate is all that is required. Simply knowing that the throughput is approximately 120 tps for one system alternative and approximately 300 tps for another alternative is sufficient information to select one option over another. These situations can be analyzed with simple performance bounding techniques presented in Chapter 3.

Detailed performance models require parameters, which can be grouped into the following categories.

• System parameters. Characteristics of a system that affect performance include, the number of processes in execution, the nature and intensity of intercomponent communication, buffer sizes, maximum number of supported threads, and network protocols.

• Resource parameters. Intrinsic features of a resource that affect performance include disk seek times, latency, transfer rates, network bandwidth, router latency, and CPU speeds.

• Workload parameters. These parameters are derived from the workload characterization and are divided into two subcategories.

  • Workload intensity parameters. provide a measure of the load placed on a system, indicated by the number of units of work that contend for system resources. Examples include the number of sessions started per day, number of transactions/day to an application server, number of operations requiring secure connections/sec, and number of database transactions executed per unit time.

  • Workload service demand parameters. specify the total amount of service time required by each basic component at each resource. Examples include the CPU time of transactions at the application server, the total transmission time of replies from the Web server back to the customer, and the total I/O time at the database server for a query function.

The critical goal of analyzing and designing IT systems is guaranteeing that performance objectives are satisfied. IT services are complex and can be very expensive. It is important to minimize the amount of guesswork when it comes to designing, implementing, and operating the systems that provide these services.