Information Technology Reference
In-Depth Information
resource co-allocation strategies are required to support not just the
management of individual resources but also the coordinated manage-
ment of multiple resources.
Grid applications typically operate on ensembles of resources that span
administrative domains of control, with resources in the ensemble being
independently operated. Furthermore, access to resources is in general
unreliable, due to either competing demands for resources or outright
failures. In a grid environment, it is not reasonable to assume a central-
ized global scheduler that resolves resource co-allocation. Two features of
grids complicate the co-allocation process, rendering ineffective existing
approaches based on centralized control and a strategy of aborting on the
failure of any resource request.
Some techniques were used to improve co-allocation effectiveness.
One approach is to enhance the local resource management system. For
example, by incorporating advance reservation capabilities into a local
resource manager, a co-allocator can obtain guarantees that a resource
will deliver a required level of service when required. Alternatively, the
resource management system can publish information about current
queue contents and scheduling policy, or publish forecasts of expected
future resource availability. This information can be used to improve the
success of co-allocation by constructing co-allocation requests that are
likely to succeed. The co-allocator may use information published by local
managers to select from among alternative candidate resources, or it may
attempt to allocate more resources than it really needs.
7. 2 . 3
Time-Critical Needs
Time-critical applications, such as medical simulation and weather
forecasting, have special requirements with respect to quality of service
(QoS) [5]. For such applications it is crucial to know at which point in time
the results of remote simulation tasks executed on some grid resources
will be available.
Let us take coastal and environmental modeling [6] as an example to
explain time-critical needs for resource allocation in a grid environment.
The economically important Louisiana Coastal Area (LCA) is one of the
world's most environmentally damaged ecosystems. Beyond the economic
loss, LCA erosion has devastating effects on its inhabitants, especially in
New Orleans whose location makes it extremely vulnerable to hurricanes
and tropical storms. To effectively model the LCA region, an integrated
framework has been developed for coastal and environmental modeling
capable of simulating all relevant interacting processes from erosion to
storm surge to ecosystem biodiversity, handling multiple time (hours to
years) and length (meters to kilometers) scales. This requires dynamically
coupled models and the invocation of algorithms based on streamed
sensor or satellite data, location of appropriate data and resources, and
 
Search WWH ::




Custom Search