This paper describes the convergence of MELCOR Accident Consequence Code System, Version 2 (MACCS2) probabilistic results of offsite consequences for the uncertainty analysis of the State-of-the-Art Reactor Consequence Analyses (SOARCA) unmitigated long-term station blackout scenario at the Peach Bottom Atomic Power Station. The consequence metrics evaluated are individual latent-cancer fatality (LCF) risk and individual early fatality risk. Consequence results are presented as conditional risk (i.e., assuming the accident occurs, risk per event) to individuals of the public as a result of the accident. In order to verify convergence for this uncertainty analysis, as recommended by the Nuclear Regulatory Commission's Advisory Committee on Reactor Safeguards, a 'high' source term from the original population of Monte Carlo runs has been selected to be used for: (1) a study of the distribution of consequence results stemming solely from epistemic uncertainty in the MACCS2 parameters (i.e., separating the effect from the source term uncertainty), and (2) a comparison between Simple Random Sampling (SRS) and Latin Hypercube Sampling (LHS) in order to validate the original results obtained with LHS. Three replicates (each using a different random seed) of size 1, 000 each using LHS and another set of three replicates of size 1, 000 using SRS are analyzed. The results show that the LCF risk results are well converged with either LHS or SRS sampling. The early fatality risk results are less well converged at radial distances beyond 2 miles, and this is expected due to the sparse data (predominance of "zero" results).
Proceedings of ExaMPI 2014: Exascale MPI 2014 - held in conjunction with SC 2014: The International Conference for High Performance Computing, Networking, Storage and Analysis
Advances in node-level architecture and interconnect technology needed to reach extreme scale necessitate a reevaluation of long-standing models of computation, in particular bulk synchronous processing. The end of Dennard-scaling and subsequent increases in CPU core counts each successive generation of general purpose processor has made the ability to leverage parallelism for communication an increasingly critical aspect for future extreme-scale application performance. But the use of massive multithreading in combination with MPI is an open research area, with many proposed approaches requiring code changes that can be unfeasible for important large legacy applications already written in MPI. This paper covers the design and initial evaluation of an extension of a massive multithreading runtime system supporting dynamic parallelism to interface with MPI to handle fine-grain parallel communication and communication-computation overlap. Our initial evaluation of the approach uses the ubiquitous stencil computation, in three dimensions, with the halo exchange as the driving example that has a demonstrated tie to real code bases. The preliminary results suggest that even for a very well-studied and balanced workload and message exchange pattern, co-scheduling work and communication tasks is effective at significant levels of decomposition using up to 131,072 cores. Furthermore, we demonstrate useful communication-computation overlap when handling blocking send and receive calls, and show evidence suggesting that we can decrease the burstiness of network traffic, with a corresponding decrease in the rate of stalls (congestion) seen on the host link and network.
Cyr, Eric C.; Chaudhry, Jehanzeb H.; Liu, Kuo; Manteuffel, Thomas A.; Olson, Luke N.; Tang, Lei
In this paper we introduce an approach that augments least-squares finite element formulations with user-specified quantities-of-interest. The method incorporates the quantity-ofinterest into the least-squares functional and inherits the global approximation properties of the standard formulation as well as increased resolution of the quantity-of-interest. We establish theoretical properties such as optimality and enhanced convergence under a set of general assumptions. Central to the approach is that it offers an element-level estimate of the error in the quantity-ofinterest. As a result, we introduce an adaptive approach that yields efficient, adaptively refined approximations. Several numerical experiments for a range of situations are presented to support the theory and highlight the effectiveness of our methodology. Notably, the results show that the new approach is effective at improving the accuracy per total computational cost.
We present PuLP, a parallel and memory-efficient graph partitioning method specifically designed to partition low-diameter networks with skewed degree distributions. Graph partitioning is an important Big Data problem because it impacts the execution time and energy efficiency of graph analytics on distributed-memory platforms. Partitioning determines the in-memory layout of a graph, which affects locality, intertask load balance, communication time, and overall memory utilization of graph analytics. A novel feature of our method PuLP (Partitioning using Label Propagation) is that it optimizes for multiple objective metrics simultaneously, while satisfying multiple partitioning constraints. Using our method, we are able to partition a web crawl with billions of edges on a single compute server in under a minute. For a collection of test graphs, we show that PuLP uses 8-39× less memory than state-of-the-art partitioners and is up to 14.5× faster, on average, than alternate approaches (with 16-way parallelism). We also achieve better partitioning quality results for the multi-objective scenario.
