A STATISTICAL LOAD DEPENDENCY MODEL FOR CPU ERRORS AT SLAC
Ravishankar K. Iyer, David J. Rossetti · 2005
EBSTRACT This paper describes an analysis of CPU errors at the Stanford Linear Accelerator Center Computational Facility. The study includes all classes of temporary and permanent CPU errors. Nearly 85 percent of the errors are temporary failures. We find a strong load dependency in the errors. The observed tendency is present in three years of load ,data. This observation is significant because a load-failure r elationship found at the CPU level must, in our vieu, be considered fundamental. In addition, the fact that most of the errors are transients or i ntermittents, provides neu information on these error types uith respect to their load dependent behavior. Our analysis procedure, used on the SLAC data, has been validated on an artificially created data base seeded uith failures.