Sammendrag
Instrumentation is ubiquitous in computer software today though its
use in parallel processing frameworks is not widespread. In this thesis,
we have developed an instrumentation framework, which we have
integrated with the P2G framework. The instrumentation framework
feeds a high and low level scheduler with detailed instrumentation data,
while inducing a minimal of overhead. Our instrumentation framework
also collects a wealth of information about the machine it runs on,
including capabilities, enabling P2G to support specialized hardware. To
demonstrate the feasibility of our framework, we have run a series of
tests that shows promising results, both for the schedulers and developers
seeking to locate a performance bottleneck, even though P2G, at the time,
was not able to use this data for enhancing the decision making process.