| dc.description.abstract | A grid computer consists of shared computing resources that are presented to the user
as a single powerful computing facility. The shared computing resources have
independent autonomous ownership. The vision in grid computing is to make
computing resources transparently available to a user without reference to the location
or ownership of the resources. The challenge is to implement grid computing in a
practical, equitable and efficient manner.
We concentrate on the scheduling aspect of this problem. The grid scheduling
problem has been approached in many ways including traditional queue based
methods, market-based methods, and agent based methods. We propose a direct
scheduling mechanism which is guided by quality of service (QoS) requirements. We
describe our experiments where the QoS requirements, memory and processor time,
guide the direct allocation of resources to users and applications using a service
matrix data structure. We present our simulation results and demonstrate that the
overall efficiency of allocations is bounded due to residual service requests or service
offers that remain unprocessed in the service matrix. The service matrix grid
scheduling methodology that we have developed addresses some of the challenges
faced in grid computing environments such as loss of scalability in centralized
control. | en |
