Enhancing performance in HPC systems under power consumption limitations

Project Title

Enhancing performance in HPC systems under power consumption limitations

Project Area of Specialization Wearables and ImplantableProject Summary

HPC technology is being shifted from Petascale to the “Exascale Computing Systems”. This powerful supercomputer will provide a thousand-fold computing power to increase over the current Petascale computing systems and will enable the unscrambling of many scientific mysteries by computing 1 exaflop (1018  calculations per second). Current supercomputers cannot deliver such a high level of computation under the power consumption limitation. Although we can achieve the Exascale level of performance by increasing the number of cores according to Exascale computing system limitations but the power consumption is another challenge.

However, our main focus is how to increase performance and decrease power consumption for future emerging technology.

Project Objectives

The main focus of the study is to enhance performance and reduce power consumption for future emerging technology. The following objectives have been set out to accomplish the research’s goal:

• Enhance System Performance through Massive Parallelism

The first objective of the current study was to enhance system performance, which is required for Exascale computing systems. Achieving an Exascale level of performance by the end of the decade will require applications to exploit billion-way parallelism provided by an envisioned Exascale system. Parallelism reduces the processing time of the process by simultaneously breaking up and running tasks on multiple microprocessors. Hence it improves performance.

• Satisfy System Power consumption by Reducing Intra-Node Communication

As new processors and memory architecture advance, clusters start to be built from systems having a large number of processors within one node to enhance the performance of HPC parallel computing systems which increase intra-node communication and hence consume more power. Putative Exascale systems will consume hundreds of megawatts using current technologies that should be limited to 20 MW according to Exascale system limitations. Therefore, reducing power consumption will be a paramount concern.

Project Implementation Method

Considering the research objectives, we have proposed a new massive parallel computing model M2C for emerging Exascale systems that provide tremendous performance through massive parallelism and limiting the overall power consumption. In order to evaluate the proposed M2C model, we implemented the proposed model in HPC benchmarking application such as Dense Matrix Multiplication (DMM).

With respect to implementations, we quantified different metrics including performance and power consumption. Under Performance we measured the fundamental parameters including time execution, speedup and number of flops whereas power consumption related metrics were including as power consumption and temperature.

All the experiments were conducted on Aziz-Fujitsu Primergy CX400 Intel Xeon True-scale QDR supercomputer containing total 11904 number of cores in the current heterogeneous system. The achieved results were better than the existing state-of-the-art technologies.

Benefits of the Project

1. The usage of high-performance computing has to turn out to be globally significant throughout the sectors of education & government and all the areas of enterprise and business.
2. High-Performance Computing provides high-end designing and simulation environment, helps applications under development to deal with marketing delivery challenges by providing the facility to accelerate or even get rid of prototyping and testing phases. And also, for the decision-making, enhancing the quality and for predicting the overall performance and failure rate of the product.
3. Moreover, HPC acts as promoting support for the research and development process in science and technology by providing support to improve existing technologies and to deliver the products to the marketplace more efficaciously and quickly.
4. Similarly, organizations and industries use supercomputers to build and examine their strategies before the actual implementation.

However, considering the manifold benefits of supercomputing it is becoming a necessity day-by-day in all fields of science and technology.

Technical Details of Final Deliverable

The final deliverable will be in the form of research i.e Research papers and the achieved results. 

Final Deliverable of the Project Software SystemType of Industry IT Technologies Wearables and ImplantablesSustainable Development Goals Industry, Innovation and InfrastructureRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	80000
Publication Fee	Equipment	2	10000	20000
Stationery and printinging	Miscellaneous	1	10000	10000
GPU device	Equipment	1	50000	50000