A structured approach to instrumentation system development and evaluation
Supercomputing '95 Proceedings of the 1995 ACM/IEEE conference on Supercomputing
Performance Evaluation of an Integrated Instrumentation System
MASCOTS '96 Proceedings of the 4th International Workshop on Modeling, Analysis, and Simulation of Computer and Telecommunications Systems
Towards scalable event tracing for high end systems
HPCC'07 Proceedings of the Third international conference on High Performance Computing and Communications
| Hi-index | 0.00 |
An instrumentation system is a measurement-based performance evaluation abstraction. Management of an instrumentation system becomes critical when tracing is enabled on an application program that runs for an excessively long period of time. This paper presents queuing models for two instrumentation system management policies, namely flush-one-when-it-fills (FOF) and flush-all-when-one-fills (FAOF). The performance results obtained analytically are compared to results obtained through trace-driven simulation. An appropriate instrumentation system management policy ensures optimal usage of the available local memory space to avoid excessive flushes of the local buffers or perturbation of the program behavior. The results show that each of the FOF and FAOF policies is suitable for one of these two objectives: the FAOF policy requires a smaller number of flushes but with greater perturbation; the converse is true for the FOF policy. Therefore, a compromise between these two policies may be needed when implementing the instrumentation system management policy.