We present a new method for reducing parallel applications' communication time by mapping their MPI tasks to processors in a way that lowers the distance messages travel and the amount of congestion in the network. Assuming geometric proximity among the tasks is a good approximation of their communication interdependence, we use a geometric partitioning algorithm to order both the tasks and the processors, assigning task parts to the corresponding processor parts. In this way, interdependent tasks are assigned to “nearby” cores in the network. We also present a number of algorithmic optimizations that exploit specific features of the network or application to further improve the quality of the mapping. We specifically address the case of sparse node allocation, where the nodes assigned to a job are not necessarily located in a contiguous block nor within close proximity to each other in the network. However, our methods generalize to contiguous allocations as well, and results are shown for both contiguous and non-contiguous allocations. We show that, for the structured finite difference mini-application MiniGhost, our mapping methods reduced communication time up to 75 percent relative to MiniGhost's default mapping on 128K cores of a Cray XK7 with sparse allocation. For the atmospheric modeling code E3SM/HOMME, our methods reduced communication time up to 31% on 16K cores of an IBM BlueGene/Q with contiguous allocation.

Many applications, such as PDE based simulations and machine learning, apply blas/lapack routines to large groups of small matrices. While existing batched blas APIs provide meaningful speedup for this problem type, a non-canonical data layout enabling cross-matrix vectorization may provide further significant speedup. In this paper, we propose a new compact data layout that interleaves matrices in blocks according to the SIMD vector length. We combine this compact data layout with a new interface to blas/lapack routines that can be used within a hierarchical parallel application. Our layout provides up to 14X, 45X, and 27X speedup against OpenMP loops around optimized dgemm, dtrsm and dgetrf kernels, respectively, on the Intel Knights Landing architecture. We discuss the compact batched blas/lapack implementations in two libraries, KokkosKernels and Intel® Math Kernel Library. We demonstrate the APIs in a line solver for coupled PDEs. Finally, we present detailed performance analysis of our kernels.

Graph algorithms are challenging to parallelize on manycore architectures due to complex data dependencies and irregular memory access. We consider the well studied problem of coloring the vertices of a graph. In many applications it is important to compute a coloring with few colors in near-lineartime. In parallel, the optimistic (speculative) coloring method by Gebremedhin and Manne is the preferred approach but it needs to be modified for manycore architectures. We discuss a range of implementation issues for this vertex-based optimistic approach. We also propose a novel edge-based optimistic approach that has more parallelism and is better suited to GPUs. We study the performance empirically on two architectures(Xeon Phi and GPU) and across many data sets (from finite element problems to social networks). Our implementation uses the Kokkos library, so it is portable across platforms. We show that on GPUs, we significantly reduce the number of colors (geometric mean 4X, but up to 48X) as compared to the widely used cuSPARSE library. In addition, our edge-based algorithm is 1.5 times faster on average than cuSPARSE, where it hasspeedups up to 139X on a circuit problem. We also show the effect of the coloring on a conjugate gradient solver using multi-colored Symmetric Gauss-Seidel method as preconditioner, the higher coloring quality found by the proposed methods reduces the overall solve time up to 33% compared to cuSPARSE.

Geometric partitioning is fast and effective for load-balancing dynamic applications, particularly those requiring geometric locality of data (particle methods, crash simulations). We present, to our knowledge, the first parallel implementation of a multidimensional-jagged geometric partitioner. In contrast to the traditional recursive coordinate bisection algorithm (RCB), which recursively bisects subdomains perpendicular to their longest dimension until the desired number of parts is obtained, our algorithm does recursive multi-section with a given number of parts in each dimension. By computing multiple cut lines concurrently and intelligently deciding when to migrate data while computing the partition, we minimize data movement compared to efficient implementations of recursive bisection. We demonstrate the algorithm's scalability and quality relative to the RCB implementation in Zoltan on both real and synthetic datasets. Our experiments show that the proposed algorithm performs and scales better than RCB in terms of run-time without degrading the load balance. Our implementation partitions 24 billion points into 65,536 parts within a few seconds and exhibits near perfect weak scaling up to 6K cores.

Considering the large number of processors and the size of the interconnection networks on exactable-capable supercomputers, mapping concurrently executable and communicating tasks of an application is complex problem that needs to be dealt with care. For parallel applications, the communication overhead can be a significant bottleneck on scalability. Topology-aware task-mapping methods that map the tasks tithe processors~(i.e., cores) by exploiting the underlying network information are very effective to avoid, or at worst bend, this limitation. We propose novel, efficient, and effective task mapping algorithms employing a graph model. The experiments show that the methods are faster than the existing approaches proposed for the same task, and on 4096 processors, the algorithms improve the communication hops and link contentions by 16% and 32%, respectively, on the average. In addition, they improve the average execution time of a parallel Spiv kernel and a communication-only application by 9% and 14%, respectively.

We present a new method for mapping applications' MPI tasks to cores of a parallel computer such that communication and execution time are reduced. We consider the case of sparse node allocation within a parallel machine, where the nodes assigned to a job are not necessarily located within a contiguous block nor within close proximity to each other in the network. The goal is to assign tasks to cores so that interdependent tasks are performed by "nearby" cores, thus lowering the distance messages must travel, the amount of congestion in the network, and the overall cost of communication. Our new method applies a geometric partitioning algorithm to both the tasks and the processors, and assigns task parts to the corresponding processor parts. We show that, for the structured finite difference mini-app Mini Ghost, our mapping method reduced execution time 34% on average on 65,536 cores of a Cray XE6. In a molecular dynamics mini-app, Mini MD, our mapping method reduced communication time by 26% on average on 6144 cores. We also compare our mapping with graph-based mappings from the LibTopoMap library and show that our mappings reduced the communication time on average by 15% in MiniGhost and 10% in MiniMD.

BACKGROUND: MicroRNAs (miRNAs) are short (19-23 nucleotides) non-coding RNAs that bind to sites in the 3'untranslated regions (3'UTR) of a targeted messenger RNA (mRNA). Binding leads to degradation of the transcript or blocked translation resulting in decreased expression of the targeted gene. Single nucleotide polymorphisms (SNPs) have been found in 3'UTRs that disrupt normal miRNA binding or introduce new binding sites and some of these have been associated with disease pathogenesis. This raises the importance of detecting miRNA targets and predicting the possible effects of SNPs on binding sites. In the last decade a number of studies have been conducted to predict the location of miRNA binding sites. However, there have been fewer algorithms published to analyze the effects of SNPs on miRNA binding. Moreover, the existing software has some shortcomings including the requirement for significant manual labor when working with huge lists of SNPs and that algorithms work only for SNPs present in databases such as dbSNP. These limitations become problematic as next-generation sequencing is leading to large numbers of novel variants in 3'UTRs. RESULT: In order to overcome these issues, we developed a web-server named mrSNP which predicts the impact of a SNP in a 3'UTR on miRNA binding. The proposed tool reduces the manual labor requirements and allows users to input any SNP that has been identified by any SNP-calling program. In testing the performance of mrSNP on SNPs experimentally validated to affect miRNA binding, mrSNP correctly identified 69% (11/16) of the SNPs disrupting binding. CONCLUSIONS: mrSNP is a highly adaptable and performing tool for predicting the effect a 3'UTR SNP will have on miRNA binding. This tool has advantages over existing algorithms because it can assess the effect of novel SNPs on miRNA binding without requiring significant hands on time.

The data one needs to cope to solve today's problems is large scale, so are the graphs and hyper graphs used to model it. Today, we have Big Data, big graphs, big matrices, and in the future, they are expected to be bigger and more complex. Many of today's algorithms will be, and some already are, expensive to run on large datasets. In this work, we analyze a set of efficient techniques to make "big data", which is modeled as a hyper graph, smaller so that its processing takes much less time. As an application use case, we take the hyper graph partitioning problem, which has been successfully used in many practical applications for various purposes including parallelization of complex and irregular applications, sparse matrix ordering, clustering, community detection, query optimization, and improving cache locality in shared-memory systems. We conduct several experiments to show that our techniques greatly reduce the cost of the partitioning process and preserve the partitioning quality. Although we only measured their performance from the partitioning point of view, we believe the proposed techniques will be beneficial also for other applications using hyper graphs.

We design, develop, and evaluate an atomic- and lock-free GPU implementation of the push-relabel algorithm in the context of finding maximum cardinality matchings in bipartite graphs. The problem has applications on computer science, scientific computing, bioinformatics, and other areas. Although the GPU parallelization of the push-relabel technique has been investigated in the context of flow algorithms, to the best of our knowledge, ours is the first study which focuses on the maximum cardinality matching. We compare the proposed algorithms with serial, multicore, and many core bipartite graph matching implementations from the literature on a large set of real-life problems. Our experiments show that the proposed pushrelabel-based GPU algorithm is faster than the existing parallel and sequential implementations.

The need to analyze high-dimension biological data is driving the development of new data mining methods. Biclustering algorithms have been successfully applied to gene expression data to discover local patterns, in which a subset of genes exhibit similar expression levels over a subset of conditions. However, it is not clear which algorithms are best suited for this task. Many algorithms have been published in the past decade, most of which have been compared only to a small number of algorithms. Surveys and comparisons exist in the literature, but because of the large number and variety of biclustering algorithms, they are quickly outdated. In this article we partially address this problem of evaluating the strengths and weaknesses of existing biclustering methods. We used the BiBench package to compare 12 algorithms, many of which were recently published or have not been extensively studied. The algorithms were tested on a suite of synthetic data sets to measure their performance on data with varying conditions, such as different bicluster models, varying noise, varying numbers of biclusters and overlapping biclusters. The algorithms were also tested on eight large gene expression data sets obtained from the Gene Expression Omnibus. Gene Ontology enrichment analysis was performed on the resulting biclusters, and the best enrichment terms are reported. Our analyses show that the biclustering method and its parameters should be selected based on the desired model, whether that model allows overlapping biclusters, and its robustness to noise. In addition, we observe that the biclustering algorithms capable of finding more than one model are more successful at capturing biologically relevant clusters.

Requirements for efficient parallelization of many complex and irregular applications can be cast as a hyper graph partitioning problem. The current-state-of-the art software libraries that provide tool support for the hyper graph partitioning problem are designed and implemented before the game-changing advancements in multi-core computing. Hence, analyzing the structure of those tools for designing multithreaded versions of the algorithms is a crucial tasks. The most successful partitioning tools are based on the multi-level approach. In this approach, a given hyper graph is coarsened to a much smaller one, a partition is obtained on the the smallest hyper graph, and that partition is projected to the original hyper graph while refining it on the intermediate hyper graphs. The coarsening operation corresponds to clustering the vertices of a hyper graph and is the most time consuming task in a multi-level partitioning tool. We present three efficient multithreaded clustering algorithms which are very suited for multi-level partitioners. We compare their performance with that of the ones currently used in today's hyper graph partitioners. We show on a large number of real life hyper graphs that our implementations, integrated into a commonly used partitioning library PaToH, achieve good speedups without reducing the clustering quality.