Computer Science Department
School of Computer Science, Carnegie Mellon University
Support for Interactive Heavyweight Services
Julio Lopez, David O'Hallaron
Keywords: Resource-aware applications, haveyweight services, active
frames, internet services, remote visualization
Interactive heavyweight services are processes that make intensive use
of resources such as computing power, memory, bandwidth and storage.
The interactive nature of these services means that they have to
satisfy user requests in a timely manner. An example of this kind of
service is the remote visualization of massive scientific datasets.
In order to present a visual representation of the dataset to the
user, the visualization process has to execute a series of compute
intensive transformation to the data. These heavyweight services
usually encounter situations of limited, heterogeneous and
dynamic resources. In order to deploy this type of service
we propose a simple mechanism, called active frames, that easily
enables us to move work and data to other compute hosts. With this
mechanism, heavyweight services can aggregate resources from multiple
compute hosts to satisfy a request. The active frame mechanism allows
the use of application-level information in the selection of resources to
satisfy the request. This is accomplished by delegating the resource
selection to an application-level scheduler. Our initial results show
that this flexible mechanism does not introduce significant overhead.