An Operational Analysis Primer Part 2: Resource-level Evaluation

An Operational Analysis Primer Part 2: Resource-level Evaluation

Author by Tom Wilson

1 Introduction

     This is part 2 of a 4-part primer on operational analysis. In part 1 ([WIL14B]), I introduced operational analysis and focused on applying it to evaluation at the system level. In this part, I will discuss evalation at the resource level. Part 3 will describe using operational analysis for modeling. Part 4 will investigate some other topics that are not found in the literature.

2 Representing System Details

     Figure 1 illustrates how a \system" can be viewed from several levels of detail. At the lowest level, a single resource can be analyzed without its queue; this is the purple server of resource 2. At the next level, a resource can be analyzed with its queue; this is the yellow box around resource 2. At another level, the entire computing system can be analyzed; this is the grey box. At the highest level, the users of the system can be analyzed along with the computing system; this is the brown box. Other levels of detail are possible, such as collections of resources (e.g., resources 2-4).

MIT 14.3 Wilson Figure 1


        Although everything can be called a "request", you might encounter other terms. At the user level, a "request" is a user . At the computing system level, a "request" might be a transaction. At the resource level, a "request" might be an operation (like read or write). They are all still "requests". Fortunately, "time" units are likely to be the same.

     In Figure 1, resource 1 has only one queue associated with a collection of resources; these resources are likely to be processors. Resources 2-4 have individual queues; these are likely to be disks or other input/output devices. The terminals do not have a queue associated with them. In this situation, it is assumed that no more users will arrive than there are terminals, or that users will leave when no terminals are available. A queue could easily be added if that reected the situation more realistically.

     In part 1, I highlighted that instrumentation to collect observed system parameters could be outside of the system. For resource parameters, the instrumentation must be inside the system. This could be an issue, but many computing systems already have some kind of instrumentation in the operating system or in other software components.

3 Parameters and Relationships

MIT 14.3 Wilson Table 1

 3.1 Observed Parameters

MIT 14.3 Wilson Figure 2 part 1

MIT 14.3 Wilson Figure 2 part 2

3.2 Derived Parameters

MIT 14.3 Wilson 3.2 Derived Parameters

 3.3 Multi-level Laws MIT 14.3 Wilson 3.3 Multi-level Laws

4 Instrumentation

MIT 14.3 Wilson 4 Instrumentation part 1

MIT 14.3 Wilson 4 Instrumentation part 2

 5 Bottlenecks

MIT 14.3 Wilson 5 Bottlenecks

 6 Conclusions

     This paper has discussed resource-level evaluation using operation analysis. Operational analysis is a well-established discipline that is very powerful. However, care should be taken when evaluating a system at multiple levels.

     The next part will investigate operational analysis applied to modeling.


MIT 14.3 Wilson Bibliography