---
_id: '18976'
abstract:
- lang: eng
  text: "Web computing is a variant of parallel computing where the idle times of
    PCs\r\ndonated by worldwide distributed users are employed to execute parallel\r\nprograms.
    In this thesis we consider a web computing variant with two\r\nimportant properties:
    First, we support the execution of coupled, massively\r\nparallel algorithms (rather
    than distributed data processing). And second,\r\nwe organize the system in peer-to-peer
    fashion.\r\n\r\nWe present the Paderborn University BSP-based Web Computing (PUB-Web)
    library,\r\nwhich supports the execution of parallel programs in the bulk-synchronous
    style\r\n(BSP) in such a web computing setting. In this thesis, we focus on important\r\ntechnical
    and algorithmic aspects, in particular: In order to schedule\r\nprocesses with
    respect to the currently available computing power, which\r\ncontinually changes
    in an unpredictable fashion, we need intelligent load\r\nbalancing algorithms
    and -- as a basic precondition -- the technical ability\r\nto migrate threads
    at runtime.\r\n\r\nTo achieve the latter in a way suitable for production use,
    compatible with\r\nrecent Java versions, available for all important platforms,
    and easy-to-use\r\nfor developers, we develop the PadMig thread migration and
    checkpointing\r\nlibrary.\r\n\r\nIn order to tackle the distributed load balancing
    problem, we present an\r\nalgorithm based on Distributed Heterogeneous Hash-Tables.
    In order to judge\r\nthe quality of the schedules produced, we perform extensive
    experiments to\r\ncompare several variants of the DHHT-based load balancer with
    the well-\r\nestablished Work Stealing algorithm, using realistic input data obtained
    by\r\nprofiling the utilization of several hundred PCs for a period of several\r\nmonths.\r\n\r\nBeside
    the available computing power, we finally also consider the network\r\nbandwidth
    as a secondary criterion for load balancing. For this purpose, we\r\ncluster the
    PUB-Web network according to bandwidth, employing a novel,\r\nfault-tolerant,
    adaptive, and scaling distributed clustering algorithm called\r\nDiDiC. In order
    to judge the quality of the clusterings produces by DiDiC,\r\nwe experimentally
    compare it to the well-established MCL algorithm using a\r\nsimulator."
author:
- first_name: Joachim
  full_name: Gehweiler, Joachim
  last_name: Gehweiler
citation:
  ama: Gehweiler J. <i>Peer-to-Peer Based Parallel Web Computing</i>. Vol 298. Verlagsschriftenreihe
    des Heinz Nixdorf Instituts, Paderborn; 2011.
  apa: Gehweiler, J. (2011). <i>Peer-to-Peer Based Parallel Web Computing</i> (Vol.
    298). Verlagsschriftenreihe des Heinz Nixdorf Instituts, Paderborn.
  bibtex: '@book{Gehweiler_2011, series={Verlagsschriftenreihe des Heinz Nixdorf Instituts,
    Paderborn}, title={Peer-to-Peer Based Parallel Web Computing}, volume={298}, publisher={Verlagsschriftenreihe
    des Heinz Nixdorf Instituts, Paderborn}, author={Gehweiler, Joachim}, year={2011},
    collection={Verlagsschriftenreihe des Heinz Nixdorf Instituts, Paderborn} }'
  chicago: Gehweiler, Joachim. <i>Peer-to-Peer Based Parallel Web Computing</i>. Vol.
    298. Verlagsschriftenreihe Des Heinz Nixdorf Instituts, Paderborn. Verlagsschriftenreihe
    des Heinz Nixdorf Instituts, Paderborn, 2011.
  ieee: J. Gehweiler, <i>Peer-to-Peer Based Parallel Web Computing</i>, vol. 298.
    Verlagsschriftenreihe des Heinz Nixdorf Instituts, Paderborn, 2011.
  mla: Gehweiler, Joachim. <i>Peer-to-Peer Based Parallel Web Computing</i>. Verlagsschriftenreihe
    des Heinz Nixdorf Instituts, Paderborn, 2011.
  short: J. Gehweiler, Peer-to-Peer Based Parallel Web Computing, Verlagsschriftenreihe
    des Heinz Nixdorf Instituts, Paderborn, 2011.
date_created: 2020-09-04T07:46:32Z
date_updated: 2022-01-06T06:53:56Z
department:
- _id: '63'
- _id: '26'
intvolume: '       298'
language:
- iso: eng
publication_identifier:
  isbn:
  - 978-3-942647-17-5
publisher: Verlagsschriftenreihe des Heinz Nixdorf Instituts, Paderborn
related_material:
  link:
  - relation: confirmation
    url: http://digital.ub.uni-paderborn.de/ubpb/urn/urn:nbn:de:hbz:466:2-186
series_title: Verlagsschriftenreihe des Heinz Nixdorf Instituts, Paderborn
status: public
title: Peer-to-Peer Based Parallel Web Computing
type: dissertation
user_id: '5786'
volume: 298
year: '2011'
...