When Is a Simple Model Adequate for Use in Scheduling in MSHN
Abstract
Current resource management systems do not provide a way to use measurements taken from an application's execution on one computer to predict that application's performance on another computer. More details are needed in both their application and resource models in order to make this prediction. However, very detailed models are also not desirable. Models that are too detailed incur unnecessary overhead when values corresponding to the detail are being obtained; they are subject to higher variances; and the benefit of computing schedules using them may be outweighed by the time required to compute those schedules. This thesis proposes a model that balances the level of detail, and therefore the quality of their predictions of resource usage, against the cost of computing schedules. To assess the quality of the proposed model, an application emulator was designed, built, and used. The results from running the application emulator demonstrated that the proposed model is able to predict the relative resource usage of an asynchronous application that has substantially more computation requirements than communication requirements. However, an even more detailed model is needed to successfully predict resource requirements of both synchronous and communication-intensive applications.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1999
- Accession Number
- ADA362793
Entities
People
- Paul F. Carff
Organizations
- Naval Postgraduate School