Allocation of SISAL program graphs to processors using BLAS Public Deposited

http://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/xg94ht02g

Descriptions

Attribute NameValues
Creator
Abstract or Summary
  • There are a number of well known techniques for extracting parallelism from a given program. They range from hardware implementations, building restructuring compilers or reorganizing of programs so as to specify all the available parallelism. The success rate of any of the known techniques is rather poor over all types of programs. This has pushed the research community to explore new languages and design different architectures to exploit program parallelism. The principles of dataflow architectures have addressed the problem of exploiting parallelism in systems by executing dataflow graphs. These graphs or programs represent data dependencies among instructions and execution of the graph proceeds in a data-driven manner. That is, an instruction is executed as soon as all its operands are available, without waiting for any program counter to sequence its execution, as is the case in conventional von Neumann architectures. In this thesis, data flow graphs are generated during the intermediate compilation of a functional language called SISAL (Streams and Iterations in a Single Assignment Language). The Intermediate Form (IFl) is a graphical language consisting of multiple acyclic function graphs that represent a given program. Each graph consists of a sequence of nodes and edges. The nodes specify the operation and the edges indicate the dependencies between the nodes. The graphs are further connected to each other by means of implicit dependencies. The Automator package developed in this project, preprocesses these multiple IF1 graphs and translates them into a single connected graph. It converts all implicit dependencies into actual ones. Additionally, complex language constructs like For All, loops and if-then-else are treated in special ways together with their nested levels by the Automator. There is virtually no limit to the number of nested levels that can be translated by this package. The Automator's prime contribution is in translating real programs written in SISAL into a specified format required by an allocation algorithm called the Balanced Layered Allocation Scheme (BLAS). BLAS partitions a connected graph into independent tasks and assigns them to processors in a multicomputer system. The problem of program allocation lies in maximizing parallelism while minimizing interprocessor communication costs. Hence, allocation is based on the best choice of communication to execution ratio for each task. BLAS utilizes heuristic rules to find a balance between computation and communication costs in the target system. Here the target architecture is a simulated nCUBE 3E computer, having a hypercube topology. Simulations show that, BLAS is effective in reducing the overall execution time of a program by considering the communication costs on the execution times. The results will help in understanding the effects in packing nodes (grain-packing), routing issues in the network and in general, the allocation problem to any processor in a network. In addition, tasks have also been assigned to adjacent processors only, instead of any processor on the hypercube network. The adjacent allocation to processors helps to determine trade-offs required between achieved speed-ups and the time it takes to completely allocate large graphs on compilation.
Resource Type
Date Available
Date Copyright
Date Issued
Degree Level
Degree Name
Degree Field
Degree Grantor
Commencement Year
Advisor
Committee Member
Academic Affiliation
Non-Academic Affiliation
Subject
Rights Statement
Peer Reviewed
Language
Digitization Specifications
  • File scanned at 300 ppi (Monochrome) using Capture Perfect 3.0 on a Canon DR-9050C in PDF format. CVista PdfCompressor 4.0 was used for pdf compression and textual OCR.
Replaces
Additional Information
  • description.provenance : Approved for entry into archive by Patricia Black(patricia.black@oregonstate.edu) on 2012-12-13T20:24:53Z (GMT) No. of bitstreams: 1 RaisinghaniManoj1994.pdf: 6693229 bytes, checksum: b7dd436e12dbbc00f28a85e099c888d1 (MD5)
  • description.provenance : Made available in DSpace on 2012-12-13T20:24:53Z (GMT). No. of bitstreams: 1 RaisinghaniManoj1994.pdf: 6693229 bytes, checksum: b7dd436e12dbbc00f28a85e099c888d1 (MD5) Previous issue date: 1994-04-07
  • description.provenance : Approved for entry into archive by Patricia Black(patricia.black@oregonstate.edu) on 2012-12-13T20:18:40Z (GMT) No. of bitstreams: 1 RaisinghaniManoj1994.pdf: 6693229 bytes, checksum: b7dd436e12dbbc00f28a85e099c888d1 (MD5)
  • description.provenance : Submitted by Kirsten Clark (kcscannerosu@gmail.com) on 2012-12-13T20:10:32Z No. of bitstreams: 1 RaisinghaniManoj1994.pdf: 6693229 bytes, checksum: b7dd436e12dbbc00f28a85e099c888d1 (MD5)

Relationships

In Administrative Set:
Last modified: 08/16/2017

Downloadable Content

Download PDF
Citations:

EndNote | Zotero | Mendeley

Items