An Intelligent Resource Sharing Mechanism for Simultaneous Multithreading CPUs Using Real Time Workload Portfolio

Date

2018

Authors

Sheikh, Marwa Naveed

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Simultaneous Multithreading improves performance of superscalar CPU's by allowing execution of multiple threads with shared path for better utilization of resources which in effect leads to improved instruction throughput by exploiting thread level parallelism. Physical Register file is one of the most critical shared resources in SMT systems due to the constraint of less number of rename registers available for renaming in order to avoid output dependencies and anti-dependencies. Due to long latency threads, the shared rename registers will not be released early and the resource becomes congested and inefficiently utilized which will result in performance degradation. In this report, an algorithm is presented by which each thread is allowed to occupy rename registers according to its requirement. The proposed method differs from the state-of-the-art such that the limit (Cap) for the rename registers for each thread is adjusted in real time according to the transient behavior of its demand and a universal lower Cap limit is adjusted autonomously based on its performance. The proposed method shows an improvement in IPC up to 53.8% in a 4 threaded system, 43.8% in a 6 threaded system and up to 41.6% in an 8 threaded system.

Description

This item is available only to currently enrolled UTSA students, faculty or staff. To download, navigate to Log In in the top right-hand corner of this screen, then select Log in with my UTSA ID.

Keywords

Citation

Department

Electrical and Computer Engineering