The Diary of a Datum: Modeling Runtime Complexity in Framework-Based
Applications

Mon Oct 24, 1 pm in CSE 1202

Nick Mitchell, Gary Sevitsky, Harini Srinivasan
IBM TJ Watson Research Center, Hawthorne, NY

What does it take to transform one field of data from one format to
another? In many of today's applications, built of layers of reusable
frameworks, simple operations like this can be surprisingly complex and
costly at run time. Our research group has been analyzing the behavior
of industrial, framework-based applications over the past several years.
It is becoming increasingly difficult to achieve good performance and to
understand behavior in these systems. In this talk we present an
approach to modeling run-time activity that makes visible the general
causes of run-time complexity and inefficiency in these systems. The
model allows us to describe and quantify behavior in terms that are
independent of the specifics of any framework, enabling comparisons
across diverse implementations. We demonstrate the approach with a
real-world case study that follows the journey of a single datum through
a tangle of frameworks. The case study also serves to illustrate the
kinds of complexity typically found in these applications