=ADD=
=reftype=
14
=number=
01-03
=url=
ftp://ftp.risc.uni-linz.ac.at/pub/techreports/2001/01-03.ps.gz
=year=
2001
=month=
01
=author=
Schreiner; Wolfgang + Kusper; Gabor + Bosa; Karoly
=title=
Introducing Fault Tolerance to Distributed Maple
=abstract=
  We have extended the parallel computer algebra environment Distributed Maple
  by fault tolerance mechanisms such that the time spent in a long running
  computation is not any more wasted by the eventual occurrence of a session
  failure.  The first mechanism is the logging of task return values and of
  shared object values such that after a failure the newly started session can
  (transparently to the application program) reuse already computed results.
  This is complicated by the fact that task arguments and results may embed
  task handles and that the scheduling layer of Distributed Maple has only a
  limited amount of information about the activities of the computing layer.
  The second mechanism is the migration of tasks such that a session may
  tolerate the failure of individual nodes without overall failure. Both fault
  tolerance mechanisms are considerably facilitated by the mostly functional
  nature of the parallel programming model; they allow to run computations
  that take much longer than the meantime between session failures.
=sponsor=
Supported by grant SFB F013/F1304 of the Austrian Science Foundation (FWF).
=keywords=
parallel computing, cluster computing, computer algebra, Java, Maple.