Information Technology Reference
In-Depth Information
response time and overhead can also be reduced if the desired result is
available in the club.
Intuitively, to set up a club requires information exchange, and clubs
need to be maintained dynamically because new GISs may join and some
existing GISs may leave. Also, it is possible that some types of services
become less popular after the club is built. Therefore, the system has to be
careful on the trade-off between the potential benei t and the cost incurred.
In general, the usage of services is not uniformly distributed. Some types
of services can be very popular and others may not. When/how clubs are
constructed/destroyed will be key issues in the S-Club scheme.
Assuming any search request is i rst sent to a GIS close to the user, on
receiving a search request for a specii c service type, the GIS checks locally
whether there has been a club for this type. If so, the GIS forwards the request
to the club, which will be l ooded within the club only. If there is no club for
this type, however, the GIS l oods the request throughout the mesh network.
When a new GIS joins the GIS network, it has no idea what clubs are
there. But since it has at least one neighbor in the underlying mesh net-
work, it can ask one of its neighbors for the information of existing clubs.
Namely, it simply copies the information of clubs from its neighbor. For a
Besides S-Club, there is an RCT (resource category tree) for the third
layer's resource management. Computational resources are usually de-
scribed by a set of attribute-value pairs. Among all attributes of a compu-
tational resource, one or several attributes are chosen to characterize the
resource capacity of meeting application resource requirements as primary
attributes (PA). An overlay called RCT (resource category tree) is used to
organize computational resources based on PAs.
Grid applications can be characterized by their requirements for
computational resources, for example, computing intensive and data-
intensive applications and, in turn, categorizing computational resources
based on certain resource characteristics that can meet application
resource requirements. By doing so, resource discovery is performed on
specii c resource categories efi ciently. For example, resources with huge
storage can better serve a data-intensive application, thus they can be
organized together based on an overlay structure.
Furthermore, according to the observation, the values of most resource
attributes are numerical, for example, values of disk size. Also, attributes
whose values are not numerical can be converted to numerical values
through certain mathematical methods. Based on this consideration, RCT
adopts an AVL tree (or balanced binary search tree) overlay structure to
organize resources with similar characteristics. The attribute that can best
describe the characteristic of resources organized by an RCT is named a
primary attribute or PA.
Figure 1.7
is an example of an RCT. The chosen
PA is available memory size, and the value domain of available memory
ranges from 0 MB to 1000 MB.
Search WWH ::
Custom Search