Modeling power and energy of the task-parallel Cholesky factorization on multicore processors

Abstract

Registro de acceso restringido Este recurso no está disponible en acceso abierto por política de la editorial. No obstante, se puede acceder al texto completo desde la Universitat Jaume I o si el usuario cuenta con suscripción. Registre d'accés restringit Aquest recurs no està disponible en accés obert per política de l'editorial. No obstant això, es pot accedir al text complet des de la Universitat Jaume I o si l'usuari compta amb subscripció. Restricted access item This item isn't open access because of publisher's policy. The full--text version is only available from Jaume I University or if the user has a running suscription to the publisher's contents.

On the Viability of Microservers for Financial Analytics

“…There is relatively little recent work performed on the measurement of energy consumption for specific numerically intensive kernels, particularly on server platforms. Alonso et al [18] model the power and energy of a specific taskparallel implementation of Cholesky factorization. Dongarra et al [19] explore the energy footprint of dense numerical linear algebra libraries on multicore systems.…”

Assessing the impact of the CPU power-saving modes on the task-parallel solution of sparse linear systems

“…We open this section by revisiting the following simple model from [10] for the total (aggregate) power dissipated by an application at a given instant of time t:…”

Modeling power and energy consumption of dense matrix factorizations on multicore processors

“…This paper makes the following specific contributions: We demonstrate that it is possible to systematically model the power and energy consumed by task‐parallel implementations of dense linear algebra operations for multicore architectures. In order to do this, we extend the results in for the Cholesky factorization, an operation involved in the solution of dense symmetric positive definite linear systems, to model two other matrix operations – the LU and QR factorizations – that are key for the solution of dense general linear systems and linear least‐squares problems.We refine the power model for the Cholesky factorization in to deliver more accurate estimations of the energy consumption. In particular, our previous model was assembled from off‐line static measures corresponding to the average power dissipation of the kernels that compose the task‐parallel algorithm for the Cholesky factorization.…”

“…In particular, our previous model was assembled from off‐line static measures corresponding to the average power dissipation of the kernels that compose the task‐parallel algorithm for the Cholesky factorization. Although we take the same initial approach, measuring power dissipation of the kernels off‐line, we now accommodate for memory contention during the actual execution that can result in high variability of power dissipation (especially as the number of cores grows) yielding a more flexible and accurate global model.We employ a power measurement device (powermeter) with a much higher sampling rate (concretely, a three‐order magnitude improvement compared with that used in ) and precision, increasing the reliability of the experimental validation of the models.…”