Cosmic microwave background map-making at the petascale and beyond

Authors:
Rajesh Sudarsan;Julian Borrill;Christopher Cantalupo;Theodore Kisner;Kamesh Madduri;Leonid Oliker;Yili Zheng;Horst Simon
Affiliations:
Lawrence Berkeley National Lab, Berkeley, CA, USA;Lawrence Berkeley National Lab, Berkeley, CA, USA;Lawrence Berkeley National Lab, Berkeley, CA, USA;Lawrence Berkeley National Lab, Berkeley, CA, USA;Lawrence Berkeley National Lab, Berkeley, CA, USA;Lawrence Berkeley National Lab, Berkeley, CA, USA;Computational Research Division, Berkeley, CA, USA;Lawrence Berkeley National Lab, Berkeley, CA, USA
Venue:
Proceedings of the international conference on Supercomputing
Year:
2011

Citing 5
Cited 1

The communication challenge for MPP: Intel Paragon and Meiko CS-2

Parallel Computing
Performance analysis of MPI collective operations

Cluster Computing
Sparse collective operations for MPI

IPDPS '09 Proceedings of the 2009 IEEE International Symposium on Parallel&Distributed Processing
MPI on a Million Processors

Proceedings of the 16th European PVM/MPI Users' Group Meeting on Recent Advances in Parallel Virtual Machine and Message Passing Interface
Scalable communication protocols for dynamic sparse data exchange

Proceedings of the 15th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming

A parallel two-level preconditioner for cosmic microwave background map-making

SC '12 Proceedings of the International Conference on High Performance Computing, Networking, Storage and Analysis

Quantified Score

Hi-index	0.00

Visualization

Abstract

The analysis of Cosmic Microwave Background (CMB) observations is a long-standing computational challenge, driven by the exponential growth in the size of the data sets being gathered. Since this growth is projected to continue for at least the next decade, it will be critical to extend the analysis algorithms and their implementations to peta-scale high performance computing (HPC) systems and beyond. The most computationally intensive part of the analysis is generating and reducing Monte Carlo realizations of an experiment's data. In this work we take the current state-of-the-art simulation and mapping software and investigate its performance when pushed to tens of thousands of cores on a range of leading HPC systems, in particular focusing on the communication bottleneck that emerges at high concurrencies. We present a new communication strategy that removes this bottleneck, allowing for CMB analyses of unprecedented scale and hence fidelity. Experimental results show a communication speedup of up to 116x using our alternative strategy.