Physical-aware task migration algorithm for dynamic thermal management of SMT multi-core processors

Abstract

-This paper presents a task migration algorithm for dynamic thermal management of SMT multi-core processors. The unique features of this algorithm include: 1) considering SMT capability of the processors for task scheduling, 2) using adaptive task migration threshold, and 3) considering cores physical features. This algorithm is evaluated on a commercial SMT quad-core processor. The experimental results indicate that our technique can significantly decrease the average and peak temperature compared to Linux standard scheduler, and two well-known thermal management techniques.

Dynamic Task Priority Scaling for Thermal Management of Multi-core Processors with Heavy Workload

“…Using the least square method, a model has been proposed to estimate the temperature and the required idle cycle injection. DTM techniques have been developed mostly for the cases light workloads (where the number of tasks is less than the number of cores) [2,4,5,6,7]. Thus, it is essential to find solutions for the cases of high workloads.…”

“…This makes a thermal difference of 10~15 ºC between the cores while running the same applications. This phenomenon is known as physical features of cores [4]. According to [4], we conducted different experiments on Intel's Core i7 3770 processor to identify the cold and hot cores.…”

“…The software techniques have lower performance overhead than hardware methods, but their effectiveness to manage and reduce the temperature is lower. Dynamic software thermal management techniques such as task scheduling [8,9,10] and task migration [2,4,5,6,9] are able to reduce the temperature with low performance overhead without additional hardware. The third method involves a combination of software and hardware techniques for dynamic thermal management [5,9].…”

“…the cases where the number of tasks is less than of number of cores [4][5][6][7]. They are not typically applicable for the heavy workloads, i.e.…”

A neuro-fuzzy fan speed controller for dynamic thermal management of multi-core processors

“…In-bound thermal management techniques operate in the critical path of the application execution. Stop-and-go and Dynamic Voltage and Frequency Scaling (DVFS) [1] are some hardware-based in-bound DTM techniques and task scheduling and task migration [2][3] are two software-based in-bound methods. On the other hand, out-ofband techniques [4] (e.g.…”