Positional adaptation of processors: application to energy reduction

Abstract

Although adaptive processors can exploit application variability to improve performance or save energy, effectively managing their adaptivity is challenging. To address this problem, we introduce a new approach to adaptivity: the Positional approach. In this approach, both the testing of configurations and the application of the chosen configurations are associated with particular code sections. This is in contrast to the currently-used Temporal approach to adaptation, where both the testing and application of configurations are tied to successive intervals in time.We propose to use subroutines as the granularity of code sections in positional adaptation. Moreover, we design three implementations of subroutine-based positional adaptation that target energy reduction in three different workload environments: embedded or specialized server, general purpose, and highly dynamic. All three implementations of positional adaptation are much more effective than temporal schemes. On average, they boost the energy savings of applications by 50% and 84% over temporal schemes in two experiments.

Characterizing and predicting program behavior and its variability

“…Prior work can be categorized into methods that define and use metrics to dynamically identify phases for adaptive optimization [5,6,8,13,26], and into techniques that identify appropriate simulation points for the desired workloads [10,16,17,24,25].…”

“…Sherwood et al [24,25] use clustering at a finer grain to classify regions of program execution into phases, and use basic block vectors as a unique signature to characterize the phases. Algorithms in [5,13] use subroutines to identify program phases and propose a hardwarebased call stack to identify major program subroutines. As our analysis will show, a simpler characterization derived from existing hardware counters is sufficient for the purpose of on-line program behavior prediction.…”

“…Several microarchitectural designs [3,5,6,8,11,13,15,18,21,31] have been proposed that allow the rapid low-cost adaptation of processor components to match hardware complexity and speed with an application's requirements. There have also been several approaches to dynamically adapt the software, such as application code, data placement, or runtime resource allocation.…”

A Detailed Study on Phase Predictors

“…Being aware of this large-scale timevarying behavior is key to understanding the behavior of the program as a whole. Phase behavior can be exploited for various purposes, ranging from performance modeling [1], compiler optimizations [2], hardware adaptation [3][4] [5][6] [7], etc. For example in phase-based hardware adaptation, if we know that particular parts of the processor are unused during some program phase, we can turn off those parts during that phase resulting in a reduced energy consumption without affecting overall performance.…”

A case for run-time adaptation in packet processing systems

“…The problem of adapting processor allocations has received considerable attention in server clusters (e.g., data centers) [4,5,8] and other domains [2,9,11,19]. However, the problem of adapting processor allocations in packet processing systems has remained virtually unexplored.…”