The scalability in terms of the time and the energy for several matrix factorizations on a multicore machine

Abstract

Scalability is an important aspect related to time and energy savings on modern multicore architectures. In this paper, we investigate and analyze scalability in terms of time and energy. We compare the execution time and consumption energy of the LU factorization (without pivoting) and Cholesky, both with Math Kernel Library (MKL) on a multicore machine. In order to save the energy of these multithreaded factorizations, the dynamic voltage and frequency scaling (DVFS) technique was used. This technique allows the clock frequency to be scaled without changing the implementation. An experimental scalability evaluation was performed on an Intel Xeon Gold multicore machine, depending on the number of threads and the clock frequency. Our test results show that scalability in terms of the execution time expressed by the Speedup metric has values close to a linear function with an increase in the number of threads. In contrast, scalability in terms of the energy consumed expressed by the Greenup metric has values close to a logarithmic function with an increase in the number of threads. Both kinds of scalability depend on the clock frequency settings and the number of threads.