In a sentence, the primary motivation for the Data Mining Grand Challenge is to provide high performance access to large datasets. This is largely an effort to apply advanced computing science tools and techniques to improve on the data access rates one would get using more conventional data I/O techniques.
Two important aspects of improving performance of any process or task is the ability to quantify the performance of the task before and after modification and the ability to identify performance bottlenecks in order to address inadequate performance.
Measuring overall performance is often fairly easy, even in a relatively complex system. This may require little more than bracketing a function call with timing statements.
However, isolating the offending code and/or component(s) responsible for poor performance can be a much more difficult undertaking in even a moderately complex system. When the complex system is also a distributed system, the difficulty of the problem increases significantly.