3.9 Workloads

< Free Open Study >

A concept we have been discussing throughout the book up to this point is that of workload, or, more simply, load. These terms represent an extremely important element in our computer systems modeling problem. The workload or load represents the events or event sequences presented to the system to model or drive the system under study. The load represents how many of some event sequence are being offered for execution during some given period of time. An example may be the number of instructions per second and the mix of instruction types presented for execution per second. The combination of the volume and the mix is important, as well as the duration of the load.

The duration may be all at once, requiring the system to queue up the requests and perform them as resources become available. This type of mix and load would saturate the system up front and then decay to no load as the items get processed. The duration could be endless, with the load continually refreshed to provide a constant saturation or equilibrium load to the system. Loads can be periodic, where the instructions are presented all at once for service and then allowed to be processed. The load is then reentered after the prescribed period of time has passed, providing another spike in processing requirement.

There is a science to workload development and selection for the computer systems modeler. For example, the database community has developed a set of transactional workloads aimed at testing a variety of database systems configurations. This set of transactions and the underlying database system have been developed over a number of years through the measurement of real database systems and the need to evaluate databases against each other with a known well-formed set of loads. Likewise, the personal computer industry has also developed a set of systems workloads aimed at allowing customers to assess the performance of one computer architecture against another.

< Free Open Study >