Paper Info
Title
Kernel-based offload of collective operations: implementation, evaluation and lessons learned
Abstract
Optimized implementations of blocking and nonblocking collective operations are most important for scalable high-performance applications. Offloading such collective operations into the communication layer can improve performance and asynchronous progression of the operations. However, it is most important that such offloading schemes remain flexible in order to support user-defined (sparse neighbor) collective communications. In this work, we describe an operating system kernel-based architecture for implementing an interpreter for the flexible Group Operation Assembly Language (GOAL) framework to offload collective communications. We describe an optimized scheme to store the schedules that define the collective operations and show an extension to profile the performance of the kernel layer. Our microbenchmarks demonstrate the effectiveness of the approach and we show performance improvements over traditional progression in user-space. We also discuss complications with the design and offloading strategies in general.
Year
DOI
Venue
2011
10.1007/978-3-642-23397-5_26
Euro-Par (2)
Keywords
Field
DocType
offloading scheme,kernel layer,performance improvement,flexible group operation assembly,kernel-based offload,traditional progression,collective operation,offloading strategy,communication layer,collective communication,asynchronous progression
Kernel (linear algebra),Asynchronous communication,Architecture,Computer science,Parallel computing,Assembly language,Implementation,Schedule,Interpreter,Distributed computing,Scalability
Conference
Volume
ISSN
Citations 
6853
0302-9743
7
PageRank 
References 
Authors
0.49
13
4
Name
Order
Citations
PageRank
Timo Schneider131218.39
Sven Eckelmann270.49
Torsten Hoefler32197163.64
Wolfgang Rehm432826.45