Energy efficient computing in grids and clouds is important for three reasons: 1) traditional HPC focuses on performance not efficiency, 2) datacenters consume large amounts of power threatening the environment, and 3) the largest supercomputers consume megawatts of power equivalent to small cities. Software solutions can improve efficiency through power-aware scheduling of tasks, data, and network communication. Researchers are estimating energy consumption of parallel applications, modeling energy requirements, and developing power-aware file systems to address this important issue.
Energy efficient computing in grids and clouds is important for three reasons: 1) traditional HPC focuses on performance not efficiency, 2) datacenters consume large amounts of power threatening the environment, and 3) the largest supercomputers consume megawatts of power equivalent to small cities. Software solutions can improve efficiency through power-aware scheduling of tasks, data, and network communication. Researchers are estimating energy consumption of parallel applications, modeling energy requirements, and developing power-aware file systems to address this important issue.
Energy efficient computing in grids and clouds is important for three reasons: 1) traditional HPC focuses on performance not efficiency, 2) datacenters consume large amounts of power threatening the environment, and 3) the largest supercomputers consume megawatts of power equivalent to small cities. Software solutions can improve efficiency through power-aware scheduling of tasks, data, and network communication. Researchers are estimating energy consumption of parallel applications, modeling energy requirements, and developing power-aware file systems to address this important issue.
Presented by presented To Ghouse Pasha VII ‘A’ & ‘B’ ISE Lecturer Dept. of Information Science & Engg. SJB Institute of Technology Bangalore Motivation Traditional research interest in cluster systems has been high performance, such as high throughput low turnaround time, load balancing Computational tasks are being created to serve the requirements of scientific community. Computational grids gives experts immediate access Comprises of geographically distributed MPPs or Clusters Largest supercomputers can consume megawatts Earth Simulator on Japan’s 5120-processor consumes 11.9 MW Enough for a city of 40,000 population Only a fraction of power consumed is used for actual computations According to Lawrence Livermore National Laboratory, for every watt of power consumed by IBM BlueGene , 0.7W of power is required to cool it. Datacenters poses great threat to environment Carbon Dioxide Emission Exponential rise in power consumption At Node level Processors are major consumers Solution for Addressing HPC suppliers are addressing this problem on a number of fronts at the chip level at the blade/box level in computing infrastructure with software Software level Solution can be power aware scheduling Task Data Blocks Communication Network Objective to improve power management and consumption using power aware ability of system devices, such as processors, disks, and communication links. Green Computing@NMIT Estimating Energy Consumed by the Parallel Application Mohan & Friends Modeling & Predicting Energy required by the parallel application Power Aware File Systems Sachin & Friends Energy Estimation E=P X N X T E: energy consumed by a program N: the number of clock cycles taken by the program T: Clock Period P: Average power consumed by a microprocessor P=I X Vcc I:Average current Vcc: Supply voltage Linux Packages (Sensors) Energy Estimation Cont.. T=1/f N: RDTSC ("read time stamp counter") instruction instruction available on recent processors Great tool for accurate timing It stores the number of elapsed clock-cycles since power-on as 64 bit integer with high 32 bits in register edx and low 32 bits in register eax. RDTSC before and after task Modelling & Prediction Predicted Energy in terms of Different No of clock cycles Different clock period Different Average power consumption Ax=E X: unknown coeffcients A is a matrix Curve-fitting technique History VM Reallocation Problem The VM reallocation problem can be divided in two: Selection of VMs to migrate Determining new placement of these VMs on physical hosts. 11 02/21/2023