Exploring Datatype Usage Space

Quantifying the use of a data structure makes benchmarking data structures easier and more reliable. We explore diierent ways of quantifying datatype usage. We present a basic solution and examine three extensions to this solution. 1 Motivation Suppose we have a selection of data structures performing similar tasks. How do we compare their eeciency? Traditionally we might choose a few benchmarks, and run each benchmark with each data structure. This has two drawbacks. Firstly, it may be hard to nd or create appropriate benchmarks. Secondly, the eeciency of a data structure may depend heavily on how it is used, though it may be unclear how a benchmark uses the data structure. However, if we could quantify use accurately, and create a benchmark for any given use, we could avoid both problems as follows. Imagine the space of possible uses of a data structure. Map out this space using a well chosen set of coordinates. Choose points at regular intervals in this space. For each point, create a benchmark with this use. Run this benchmark with each competing data structure. We now have a table of eeciencies qualiied by a broad selection of uses. See Fig.1. Suppose now we have an application using any one of these data structures. By nding how this application uses the data structure, we can identify a point in the usage space. By examining the nearest point in our results table , we can estimate the performance of the competing data structures for this application|see Fig.1 again. This estimation will in general be more reliable than that obtained using hand-picked benchmarks. If the application changes how it uses the data structure|because we change the input data or the program itself|we can justify any change in data structure with knowledge about the change in use. We need to deene the use of a data structure: Section 2 describes a model capturing datatype usage. Section 3 deenes a basic set of coordinates for the datatype usage space. Section 4 examines principles for mapping out the data-type usage space. Section 5 presents three possible extensions to the basic set of coordinates. Section 6 refers to related work. Section 7 concludes. 2 Datatype Usage Graph We want to capture the datatype usage of an application. The application can use one of many competing data structures. To abstract away from the choice of