10.6 Data presentation

< Free Open Study >

If one cannot prepare and present the results of a performance study clearly and simply, then the study would be deemed a failure, no matter how much effort was put into the work. The aim of every performance study is to aid the analyst and associated client in making a decision regarding the computer system being studied. To aid in this analysis the modeler must possess the ability to determine what medium is best to use in making specific information available-for example, words, graphs, pictures, charts, animation, or some other means amenable to the domain being studied.

The old saying that a picture is worth a thousand words is one the modeler must take to heart and strive to realize. Graphics are one of the best means to convey differences between studied components or systems. It is relatively easy to see that one CPU performs better than another when they are shown clearly in graphical form and the graph clearly depicts the relative performance differences. There are many kinds of graphics available to depict such comparisons-for example, line graphs, bar charts, pie charts, histograms, and Gantt charts. In all cases it is critical that we understand what is being plotted and why, in order that we select the correct variables and styles in which to represent them.

One such value that impacts the choice of which chart to use is the type of variable displayed. Is the variable being plotted quantitative or qualitative, is it ordered or unordered, is it discrete, or is the value continuous? Qualitative variables are those where there is no specific measure present, merely a category. For example, microprocessors, servers, and mainframe computers are all classes of computers, but there is no measured value when we use these terms alone. Quantitative values are those that we can measure explicitly-for example, the number of instructions per second or the number of I/O requests per period. We would probably use a line graph to show the time-based relationship between a continuous set of variables. On the other hand, if we had discrete value variables, we may decide to use a histogram or bar chart to depict these.

When deciding what form to use it is important that the modeler keep a few important concepts in mind. First, choose a reporting mechanism that will require minimum effort from the reader. The differences you wish to depict should be clearly defined and displayed so that the client will have no problem coming to the same conclusion that the modeler did after the experiments were run. Make sure that all pertinent information is provided on the graph so the reader need not look elsewhere to fill in the blanks. Keep it simple. Even though the second item indicated to put all pertinent information on the graph, one also must make sure that no nonuseful information finds its way onto the graph. Try to use standard methods of describing and displaying information. For example, the origin of the graph is expected by most people to be labeled as the zero point in both dimensions. Finally, try to avoid ambiguity. For example, make sure all axes are labeled clearly (e.g., use names, not symbols), show the scales used clearly (e.g., log scales, decimal, etc.), and use clear differences to depict different values of variables (e.g., CPU type 1 is red, CPU type 2 is black, etc.). The scales being used should be set so they clearly depict the differences. This last important concept should not be overlooked. Choosing an inappropriate scale may make a claim look better or worse than it really is. The interested reader is pointed to texts on statistics that focus on data representation for more complete discussions and examples of some of these concepts.

< Free Open Study >