Article Preview
Top1. Introduction
High-Performance Computing (HPC) refers to the system that performs parallel operations very fast and produces answers or feedback within a short time (Kohlmeyer, 2014). HPC is becoming a more important system in many research disciplines in solving complex problems such as weather forecasting, molecular modeling of biological compounds, simulation, data mining, and visualization (Kohlmeyer, 2013). However, the data center is consuming an increasing amount of energy in which a large part of the consumption is generated by HPC systems as described in (Onime, 2013 & Mammela, 2011) that HPC is high power computing and high energy consumption respectively. Therefore, data centers and ICT industries are consuming an increasing amount of energy, and a large part of the consumption is generated by the presence of HPC facilities. This makes the power control required by HPC facilities to become a key challenge for the effective operation of modern high-end computing infrastructure (Surve et al., 2013).
The increase in the energy consumption of data centers has become a critical issue despite the growing demand for higher performance-computing infrastructure (Garg et. al., 2009). For example, in the year 2019, one of the world's fastest supercomputer leading number one between 2018 and 2019, IBM Power System AC922, IBM POWER9 22C 3.07GHz, NVIDIA Volta GV100 of Ridge National Laboratory from the United States of America consumed 10.096 MW of power (Top500, 2019) while the same machine on 2018 consumed 9.783 MW (Top500, 2018) which implied the increases of power consumption by 3.2%. Similarly, another one from Japan (HA8000-tc HT210/PRIMERGY CX400 Cluster) consumed 19.4313MW (Top500, 2017). The average annual energy consumption in the sub-Saharan domestic sector is 488-kilowatt hours (kWh) per capita which is nearly equivalent to 5% of the consumption in the United States (Avila et al., 2017).
Correspondingly, the HPC of Dar es Salaam Institute of Technology (DIT) in the year 2017 required a total power of 38.7kW, while in 2015, it required only 19.6kW. It is improbable that supercomputing centers can continue increasingly utilizing resources is proportional to the increase of computing powers. In particular, lowering energy consumption can save money while increasing reliability (Freeh et al., 2007). Therefore, with this cost, if energy consumption will not be reduced, similar institutions that have deployed HPC systems will face an increase in budget in settling electricity bills. It should be noted that high-energy consumption is a result of high power consumed caused by high power computing. Generally, this is a result of more computational jobs that require high power and high network bandwidth. Communication increases as well due to data movement regardless of the need for high bandwidth (Jin et al., 2016).
This study proposes a Hybrid Schedule Algorithm that combines techniques from priority and shortest remain time first (SRTF) algorithms. This will improve the energy efficiency of high-performance computing which is the main goal of this study. The term High-Performance Computing system (HPC) used as a synonym for Supercomputer refers to the computer with a high level of performance as compared to a general-purpose computer (Wikipedia, 2020).