Run-time Support for Adaptive Heavyweight Services

		Julio Lopez and David O'Hallaron
		   {jclopez,droh}@cs.cmu.edu

		   Carnegie Mellon University

By definition, a heavyweight network service requires a significant amount
of computation to complete its task.  Providing a heavyweight service is
challenging for a number of reasons.  First, since the service can
typically not be provided in a timely fashion using a single server at the
remote site, multiple hosts at both the server and client sites must be
employed. Second, the available compute and network resources change with
respect to time.  Thus, an effective service must be adaptive in the sense
that is able to transparently aggregate the available resources and react
to the changing availability of these resources.  In this paper we present
a framework that allows us to build these kinds of adaptive heavyweight
services.  Experimental results with a distributed visualization service
suggest that the cost imposed by the new capability is reasonable.


@incollection	{ lopez-lcr00,
  author	= "Julio Lopez and David O'Hallaron",
  title		= "Runtime Support for Adaptive Heavyweight Services",
  pages		= "221--234",
  series	= "Lecture Notes in Computer Science",
  booktitle	= "Proc. of 5th Workshop on Languages, Compilers, and
                   Run-time Systems for Scalable Computers ({LCR}'2000)",
  volume	= 1915,
  year		= 2000,
  address	= "Rochester, NY",
  publisher	= "Springer--Verlag"
}