Page 1 of 15

Transactions on Engineering and Computing Sciences - Vol. 12, No. 6

Publication Date: December 25, 2024

DOI:10.14738/tecs.126.18051.

Agrawal, M. & Mehta, D. A. (2024). Thermal Aware Process Scheduling for Multicore Processors. Transactions on Engineering and

Computing Sciences, 12(6). 39-53.

Services for Science and Education – United Kingdom

Thermal Aware Process Scheduling for Multicore Processors

Mahima Agrawal

Computer Engineering Department,

S G S Institute of Technology and Science, Indore, M.P., India

D. A. Mehta

Computer Engineering Department,

S G S Institute of Technology and Science, Indore, M.P., India

ABSTRACT

Multi-core processors seem to be an alternative way to higher frequencies for

increasing microprocessor performance, by handling more work in parallel at

lower frequencies. The addition of multiple cores on the same chip results in an

increase in power density on the chip which in turn generates large amount of heat.

The increased temperature increases leakage current; and negatively affects chip’s

performance, reliability and life expectancy. In addition, they release greenhouse

gases in the atmosphere and have negative impact on the environment. The study

done so far reveals that the issue of high temperature can be solved by assigning

and migrating the processes to a cooler core; but this increases migration cost and

temporal temperature gradients thereby decreasing the performance. So, in this

work the issue of high temperature along with temporal temperature gradients of

the cores is addressed at the Operating System level via scheduling of processes.

The experiments are performed on an Intel i5-3470 Linux machine. The

experimental results reveal reduction in the peak temperature of the processor up

to 11.36%, thermal swings in the processor up to 80% and turnaround time of the

processes up to 17.24%. The idea of thermal aware scheduler presented in this

paper can be applied for scheduling of jobs in data centers and high-performance

computers to achieve performance while making computing environmentally

friendly.

Keywords: Completely Fair Scheduler, Green Computing, Linux, Reliability, Scheduling,

Thermal metric.

INTRODUCTION

Performance was the ultimate goal of the IT designers until there was no need for energy-aware

computing. Earlier high performance was achieved by increasing the frequency of processors.

Later, it was discovered that processors running at high frequency causes increased power

consumption. This led to the invention of multi-core processors where two or more cores reside

on the same physical chip. Though multi-core processors run at low frequency but increases

the transistor count on the chip. With each successive generation, this increasing number of

transistors increased the power consumption which in turn, with the decrease in chip size

increased the power density (i.e., the power consumed per unit area of the chip). The power

density increases approximately by a factor of k2 every generation [1]. Also, Shekhar Borkar an

Intel Fellow says, 'with leakage power dominating, power consumption is roughly proportional

Page 2 of 15

40

Transactions on Engineering and Computing Sciences (TECS) Vol 12, Issue 6, December - 2024

Services for Science and Education – United Kingdom

to transistor count' [2]. Intel has already yielded two experimental chips of 48 and 80 cores

through its Terascale Computing Research Program and is supposed to integrate 256 billion

transistors on a feature size of 8nm through its Exascale Computing Research Program.

Therefore, multi-core systems will continue to face high power densities. The exponential

increase in power density will result in increase in temperature. Moreover, diversity of

applications also contributes in increase in power and temperature as applications usually have

different thermal intensity. Different applications running on a system may contribute

differently to the overall temperature, depending on the way they utilize the processor

resources. For instance, while doing typical word processing, email tasks, the processor is idle

most of the time and will run very cool whereas while running complicated tasks like 3-D

rendering, games, the processor is more active and heat up frequently.

Each device incorporates a processor to deliver performance whilst generating greater amount

of heat thus increasing CO2 emissions. These CO2 emissions adversely affect health and

contribute towards global climate changes. As the temperature of the processor increases, there

is an increase in leakage current which in turn increases the temperature again leading to

thermal runaway. Also, when the temperature of the integrated circuit in processor rises

beyond a specified rate, the operating behavior of the circuit changes thus degrading

performance. In addition, the likelihood of catastrophic failure (through mechanisms such as

electro-migration, junction fatigue, gate oxide breakdown, thermal runaway, package

meltdown, etc.) also increases exponentially thus negatively affecting reliability. The negative

effect of high temperature is not only limited to chip's reliability and life expectancy but also to

smaller mobile devices with limited battery capacity and to ICT (Information and

Communication Technology). Further, the growing use of Internet, Web Applications, E- commerce and Cloud Computing to store and retrieve information digitally requires large

number of servers and data centers. These servers and data centers have higher performance

power demands which leads to increase in power density, high temperature and cooling

mechanism; thereby increasing overall operational costs and raising environmental concerns

like increasing CO2 emissions. Moreover, reliability of hardware is an important aspect of green

computing because it reduces the overall costs of energy associated with system failures and E- waste. Increasing the reliability of the IT infrastructure leads to significant energy savings while

minimizing hazardous waste materials and their disposal. [3] With the growing computing

needs, restrictions on energy supply and access, and global climate changes, it is the need of

hour to manufacture, use, design and recycle the IT equipments efficiently and effectively with

minimal or no impact on the environment. So, in the current scenario it is extremely important

to design an optimal global solution to address the issues of saving energy, reducing cooling

costs, increasing performance and reliability of the processor. However, technically the major

temperature induced problems for multi-core processor are the absolute rise in temperature,

the spatial temperature gradient and the temporal temperature gradient.

The rise in processor temperature can damage the processor components owing to thermal

runaway, thereby reducing the reliability of the processor. To reduce thermal resistance and

improve heat flow away from the chip, the hardware designers used solutions like heat sinks,

air cooling, liquid-cooling, phase-change cooling, efficient floor planning. However, as the chip

size continues to shrink, the integration of these cooling techniques becomes expensive. The