2017 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)最新文献

筛选
英文 中文
Snowflake: A Lightweight Portable Stencil DSL 雪花:一个轻量级的可移植模板DSL
Nathan Zhang, Michael B. Driscoll, Chick Markley, Samuel Williams, P. Basu, A. Fox
{"title":"Snowflake: A Lightweight Portable Stencil DSL","authors":"Nathan Zhang, Michael B. Driscoll, Chick Markley, Samuel Williams, P. Basu, A. Fox","doi":"10.1109/IPDPSW.2017.89","DOIUrl":"https://doi.org/10.1109/IPDPSW.2017.89","url":null,"abstract":"Stencil computations are not well optimized by general-purpose production compilers and the increased use of multicore, manycore, and accelerator-based systems makes the optimization problem even more challenging. In this paper we present Snowflake, a Domain Specific Language (DSL) for stencils that uses a \"micro-compiler\" approach, i.e., small, focused, domain-specific code generators. The approach is similar to that used in image processing stencils, but Snowflake handles the much more complex stencils that arise in scientific computing, including complex boundary conditions, higher-order operators (larger stencils), higher dimensions, variable coefficients, non-unit-stride iteration spaces, and multiple input or output meshes. Snowflake is embedded in the Python language, allowing it to interoperate with popular scientific tools like SciPy and iPython; it also takes advantage of built-in Python libraries for powerful dependence analysis as part of a just-in-time compiler. We demonstrate the power of the Snowflake language and the micro-compiler approach with a complex scientific benchmark, HPGMG, that exercises the generality of stencil support in Snowflake. By generating OpenMP comparable to, and OpenCL within a factor of 2x of hand-optimized HPGMG, Snowflake demonstrates that a micro-compiler can support diverse processor architectures and is performance-competitive whilst preserving a high-level Python implementation.","PeriodicalId":396022,"journal":{"name":"2017 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-06-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123412637","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 13
A Locality-Based Threading Algorithm for the Configuration-Interaction Method 一种基于位置的配置交互线程算法
H. Shan, Samuel Williams, C. Johnson, K. McElvain
{"title":"A Locality-Based Threading Algorithm for the Configuration-Interaction Method","authors":"H. Shan, Samuel Williams, C. Johnson, K. McElvain","doi":"10.1109/IPDPSW.2017.15","DOIUrl":"https://doi.org/10.1109/IPDPSW.2017.15","url":null,"abstract":"The Configuration Interaction (CI) method has been widely used to solve the non-relativistic many-body Schrodinger equation. One great challenge to implementing it efficiently on manycore architectures is its immense memory and data movement requirements. To address this issue, within each node, we exploit a hybrid MPI+OpenMP programming model in lieu of the traditional flat MPI programming model. In this paper, we develop optimizations that partition the workloads among OpenMP threads based on data locality,-which is essential in ensuring applications with complex data access patterns scale well on manycore architectures. The new algorithm scales to 256 threadson the 64-core Intel Knights Landing (KNL) manycore processor and 24 threads on dual-socket Ivy Bridge (Xeon) nodes. Compared with the original implementation, the performance has been improved by up to 7× on theKnights Landing processor and 3× on the dual-socket Ivy Bridge node.","PeriodicalId":396022,"journal":{"name":"2017 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)","volume":"37 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-06-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131682648","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 3
Efficient Anytime Anywhere Algorithms for Vertex Additions in Large and Dynamic Graphs 高效的随时随地算法在大型和动态图顶点添加
Eunice E. Santos, John Korah, Vairavan Murugappan, Suresh Subramanian
{"title":"Efficient Anytime Anywhere Algorithms for Vertex Additions in Large and Dynamic Graphs","authors":"Eunice E. Santos, John Korah, Vairavan Murugappan, Suresh Subramanian","doi":"10.1109/IPDPSW.2017.179","DOIUrl":"https://doi.org/10.1109/IPDPSW.2017.179","url":null,"abstract":"Over the past decade, there has been a dramatic increase in the availability of large and dynamic social network datasets. Conducting social network analysis (SNA) on these networks is critical for understanding underlying social phenomena. However, continuously evolving graph structures require massive recomputations and conducting SNA is infeasible if the computations have to be restarted for every change. Many recent works have proposed large-scale graph processing systems and frameworks, but less attention has been given to scalable SNA algorithm designs that can efficiently adapt to dynamic graph changes. Moreover, continuously adapting to dynamic graph changes such as node/vertex/actor additions/deletions in a parallel/distributed computational environment can skew the initial graph partitions, leading to load imbalance issues and performance degradation. Previous approaches that focus on computing SNA measures on dynamic graphs either ignore this critical load-balancing aspect or focus only on measures that are straightforward and inherently adjustable to changes in the graph topology. In this work, we have designed an anytime anywhere closeness centrality algorithm that can efficiently incorporate vertex additions while avoiding massive recomputations, by leveraging a generic framework for designing parallel/distributed algorithms called anytime anywhere. Furthermore, we have also performed an analysis of the effectiveness of various processor assignment strategies to mitigate the load imbalances caused by dynamic graph changes.","PeriodicalId":396022,"journal":{"name":"2017 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)","volume":"130 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-06-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122157632","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 2
Towards Trustworthy Testbeds Thanks to Throughout Testing 通过全面的测试,迈向值得信赖的测试平台
L. Nussbaum
{"title":"Towards Trustworthy Testbeds Thanks to Throughout Testing","authors":"L. Nussbaum","doi":"10.1109/IPDPSW.2017.101","DOIUrl":"https://doi.org/10.1109/IPDPSW.2017.101","url":null,"abstract":"When using testbeds in the context of experimental computer science, theability to produce trustworthy and reproducible experiments results dependsgreatly on the trustworthiness of the infrastructure itself. Unfortunately, several factors many issues such as software misconfiguration, hardwareheterogeneity, or service failures, can remain undetected and affect thequality of experimental results. This paper presents the design andimplementation of an automated testbed testing framework. This framework wasdeployed in the context of the Grid'5000 project, and uncovered more than onehundred of issues.","PeriodicalId":396022,"journal":{"name":"2017 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)","volume":"17 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-06-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115762790","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 5
Minimizing I/Os in Out-of-Core Task Tree Scheduling 最小化核外任务树调度的I/ o
L. Marchal, Samuel McCauley, B. Simon, F. Vivien
{"title":"Minimizing I/Os in Out-of-Core Task Tree Scheduling","authors":"L. Marchal, Samuel McCauley, B. Simon, F. Vivien","doi":"10.1109/IPDPSW.2017.58","DOIUrl":"https://doi.org/10.1109/IPDPSW.2017.58","url":null,"abstract":"Scientific applications are usually described as directed acyclic graphs, where nodes represent tasks and edges represent dependencies between tasks. For some applications, such as the multifrontal method of sparse matrix factorization, this graph is a tree: each task produces a single output data, used by a single task (its parent in the tree). We focus on the case when the data manipulated by tasks have a large size, which is especially the case in the multifrontal method. To process a task, both its inputs and its output must fit in the main memory. Moreover, output results of tasks have to be stored between their production and their use by the parent task. It may therefore happen, during an execution, that not all data fit together in memory. In particular, this is the case if the total available memory is smaller than the minimum memory required to process the whole tree. In such a case, some data have to be temporarily written to disk and read afterwards. These Input/Output (I/O) operations are very expensive; hence, the need to minimize them. We revisit this open problem in this paper. Specifically, our goal is to minimize the total volume of I/O while processing a given task tree. We first formalize and generalize known results, then prove that existing solutions can be arbitrarily worse than optimal. Finally, we propose a novel heuristic algorithm, based on the optimal tree traversal for memory minimization. We demonstrate good performance of this new heuristic through simulations on both synthetic trees and realistic trees built from actual sparse matrices.","PeriodicalId":396022,"journal":{"name":"2017 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)","volume":"14 6","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"113967808","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 3
Efficient Data Structures for a Hybrid Parallel and Vectorized Particle-in-Cell Code 并行和向量化混合粒子单元码的高效数据结构
Yann Barsamian, S. Hirstoaga, E. Violard
{"title":"Efficient Data Structures for a Hybrid Parallel and Vectorized Particle-in-Cell Code","authors":"Yann Barsamian, S. Hirstoaga, E. Violard","doi":"10.1109/IPDPSW.2017.74","DOIUrl":"https://doi.org/10.1109/IPDPSW.2017.74","url":null,"abstract":"The contribution of the present work relies on an innovative and judicious combination of several optimization techniques for achieving high performance when using automatic vectorization and hybrid MPI/OpenMP parallelism in a Particle-in-Cell (PIC) code. The domain of application is plasma physics: the code simulates 2d2v Vlasov-Poisson systems on Cartesian grids with periodic boundary conditions. Overall, our code processes 65 million particles/second per core on Intel Haswell (without hyper-threading) and achieves a good weak scaling up to 0.4 trillion particles on 8,192 cores. The optimizations mainly consist in using (i) a structure of arrays for the particles, (ii) an efficient data structure for the electric field and the charge density, and (iii) an appropriate code for automatic vectorization of the charge accumulation and of the positions’ update. In particular, we use space-filling curves to enhance data locality while enabling vectorization: starting from a redundant cell-based data structure for the electric field and for the charge density, we compare several space-filling curves for an efficient ordering of these data and we obtain a gain of 36% in the number of L2 and L3 cache misses when using a Morton curve instead of the classical rowmajor one. In addition, by proposing a specific writing of the updating positions code we achieve a 31% time improvement in that step. The optimizations bring an overall gain in the execution time of 42% with respect to a standard code. The parallelization of the particle loops is simply performed by means of both distributed and shared memory paradigms, without domain decomposition. We explain the weak and the strong scalings of the code bounded as expected by the overhead of the MPI communications.","PeriodicalId":396022,"journal":{"name":"2017 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)","volume":"179 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115566480","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 6
ReEP: A Toolset for Generation and Programming of Reconfigurable Datapaths for Event Processing ReEP:用于事件处理的可重构数据路径生成和编程的工具集
P. Gottschling, C. Hochberger
{"title":"ReEP: A Toolset for Generation and Programming of Reconfigurable Datapaths for Event Processing","authors":"P. Gottschling, C. Hochberger","doi":"10.1109/IPDPSW.2017.138","DOIUrl":"https://doi.org/10.1109/IPDPSW.2017.138","url":null,"abstract":"Reconfigurable datapaths can be used to implement multiple applications on the same hardware. Switching between applications can be realized by loading new configuration information into the datapath. In this contribution, we want to use such datapaths for high frequency event processing. We have developed the toolset ReEP, which takes multiple problem descriptions and superposes them into one reconfigurable datapath. This generalized datapath can be synthesized and programmed onto an FPGA, but could as well be used on an ASIC. By loading a configuration into the datapath it is turned into a problem specific datapath. ReEP is further able to map arbitrary problems to such pre-synthesized datapaths - given that all required resources are available - and calculates a configuration for the given problem. The necessary time for mapping, creating a configuration and downloading it to the reconfigurable datapath is orders of magnitude faster than dynamic partial reconfiguration.","PeriodicalId":396022,"journal":{"name":"2017 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)","volume":"96 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123154290","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 3
A Near Optimal Integrated Solution for Resource Constrained Scheduling, Binding and Routing on CGRAs CGRAs上资源约束调度、绑定和路由的近最优集成解
Tajas Ruschke, L. Jung, C. Hochberger
{"title":"A Near Optimal Integrated Solution for Resource Constrained Scheduling, Binding and Routing on CGRAs","authors":"Tajas Ruschke, L. Jung, C. Hochberger","doi":"10.1109/IPDPSW.2017.99","DOIUrl":"https://doi.org/10.1109/IPDPSW.2017.99","url":null,"abstract":"Mapping problems to Coarse Grained Reconfigurable Arrays (CGRA) has been researched for many years now. Yet, no feasible mapping algorithms are known that can be considered optimal or even near optimal. The main reason for this deficit is the complex nature of the mapping problem. It can be considered as a combined scheduling, binding and routing problem. It involves several constraints that need to be handled at the same time: operator availability, storage space and interconnect. This becomes even more distinct if we consider inhomogeneous and irregular CGRAs. In this contribution we establish lower bounds for schedule lengths by employing the known optimal resource constrained scheduling approach BULB. We modify it such that it can handle the additional constraints of CGRAs. Although the runtime can get very long, this modified BULB scheduler can find near optimal schedules for all test cases. It is also able to handle real life applications like an ADPCM decoder consisting of more than 600 operations.","PeriodicalId":396022,"journal":{"name":"2017 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)","volume":"48 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133217452","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 5
Communication Aware task Placement for Workflow Scheduling on DaaS-Based Cloud 基于daas云的工作流调度的通信感知任务布局
Hadrien Croubois, E. Caron
{"title":"Communication Aware task Placement for Workflow Scheduling on DaaS-Based Cloud","authors":"Hadrien Croubois, E. Caron","doi":"10.1109/IPDPSW.2017.96","DOIUrl":"https://doi.org/10.1109/IPDPSW.2017.96","url":null,"abstract":"Cloud platforms have emerged as a leading solution for computation. In the meantime, large computations have shifted from big parallel tasks to workflows of smaller tasks with data dependencies between them. Task placement is a major issue on Cloud platforms, especially considering the impact of data exchanges on cost and makespan. In this paper, we investigate the consequences of network contention regarding the use of existing scheduling policies on DaaS-based platforms (DaaS for Data as a Service). We show here that the legacy algorithms use inefficient network models. We then modify those algorithms using a new model inspired by DaaS-based Cloud platforms. Thus, we manage to statically pack tasks so that a batch scheduler could deploy many real-time submitted workflows on a dynamic Cloud platform. Simulations of Fork-Join workflows deployment using SimGrid show that our algorithm reduces computation time as well as deployment costs.","PeriodicalId":396022,"journal":{"name":"2017 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)","volume":"53 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130813361","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 4
Towards Reproducible Blocked LU Factorization 面向可复制的阻塞LU分解
R. Iakymchuk, E. S. Quintana‐Ortí, E. Laure, S. Graillat
{"title":"Towards Reproducible Blocked LU Factorization","authors":"R. Iakymchuk, E. S. Quintana‐Ortí, E. Laure, S. Graillat","doi":"10.1109/IPDPSW.2017.94","DOIUrl":"https://doi.org/10.1109/IPDPSW.2017.94","url":null,"abstract":"In this article, we address the problem of reproducibility of the blocked LU factorization on GPUs due to cancellations and rounding errors when dealing with floating-point arithmetic. Thanks to the hierarchical structure of linear algebra libraries, the computations carried within this operation can be expressed in terms of the Level-3 BLAS routines as well as the unblocked variant of the factorization, while the latter is correspondingly built upon the Level-1/2 BLAS kernels. In addition, we strengthen numerical stability of the blocked LU factorization via partial row pivoting. Therefore, we propose a double-layer bottom-up approach for ensuring reproducibility of the blocked LUfactorization and provide experimental results for its underlying blocks.","PeriodicalId":396022,"journal":{"name":"2017 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133488086","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
0
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
相关产品
×
本文献相关产品
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:604180095
Book学术官方微信
小红书