Optimal performance state estimation of compute systems

This paper presents a novel approach to estimating and predicting the system-wide utilisation of computational re-sources in real-time. An algorithm that implements a dis-crete minimum mean-square error lter is applied to fuse concurrent and sequential observations of system event counts into a state vector. Contemporary computer com-ponents and subsystems make these event counts avail-able through hardware performance counter registers. The registers may be accessed by the system's software quasi-concurrently but the number of registers in individual com-ponents is usually smaller than the number of events that can be monitored. Our approach overcomes this prob-lem by modeling individual hardware performance counter readings as vector random processes and recursively pro-cesses them one at a time into a common state vector, thereby making larger performance counter sets observ-able than would otherwise be possible. 1