Cluster level modeling and analysis

The objective of this track is to provide the proper infrastructure to model and analyze the whole cluster, and to support building special analysis modules that examine different aspects of the computing load. One Postdoctoral student will work on cluster level modeling and efficient storage of the tracing, sampling and model data. The other Postdoctoral student will work on the automated optimization of the data collection framework. He will propose new algorithms to optimize the sampling frequency (of performance counters or state variables in processes) and the set of activated tracing probes, based on the system load and capacity, and according to the desired level of accuracy and acceptable overhead. As such, the algorithms will need to predict the associated overhead, given the system model, of changing the sampling frequency or set of activated probes.




Documents and presentations

Model-based constraints over execution traces to analyze multi-core and real-time systems

.(Not yet published)

Detection of common problems in real-time and multi-core systems using model-based constraints

.(Not yet published)

Execution path profiling using hardware performance counters

.(Not yet published)

Precisely trace a request in Cloud environment

Distributed critical path extraction from kernel trace

Virtual machine monitoring using trace analysis

TMF May 2013

Efficient analysis of application servers in the Cloud