Grid Resource Allocation - Grid Computing: Infrastructure, Service, and Applications - page 179

FIGURE 7.6

HARC architecture and resource reservation procedure.

Although a local resource manager can be used for local reservation, it
is inconvenient to manually reserve geographically distributed resources
for applications that might simultaneously have access to those distri-
buted resources. HARC (the highly available resource co-allocator [30])
was developed to reserve multiple types of resources in a single step. It
reserves multiple resources in an all-or-nothing fashion (i.e., atomically)
and provides a co-allocation service supporting both metacomputing and
workl ow applications.
The HARC architecture, shown in Figure 7.6, consists of clients (request-
ing resource co-allocation), acceptors (making reservations), and resource
managers (talking with local schedulers). The clients i rst contact the
resource manager (RM) for timetable information. The timetable gives
information such as what resource is free and when it is free, and so on.
The client analyzes the information, selects a set of resources, and sends
the requests to acceptors (Acc) asking for co-scheduling. The Acc sends
back the transaction identii er (TID) to clients to differentiate the
co-allocation.
Then Acc sends Prepare messages to all the RMs asking for scheduling
of the resources based on a client's request. Each RM evaluates the request
and responds with a Prepared message if it can meet the requirement.
However, since the timetable will be slightly out of date, there is a pos-
sibility that the request will be Aborted . Acc collects all the responses from
RMs, returns a Commit message if all RMs can meet the requirements, or