Grid Resource Allocation - Grid Computing: Infrastructure, Service, and Applications - page 170

7. 3 . 2 Phase 2: System Selection

Given a group of possible resources (or a group of possible resource sets),
all of which meet the minimum requirements for the job, a single resource
(or single resource set) must be selected on which to schedule the job. This
selection is generally done in two steps: gathering detailed information
and making a decision. In order to make the best possible job/resource
match, detailed dynamic information about the resources is needed. Since
this information may vary with respect to the application being sched-
uled and the resources being examined, no single solution will work in
all, or even most, settings. The dynamic information gathering step has
two components: what information is available and how the user can
access it. The available information will vary from site to site, and users
currently have two main sources: the grid information service (GIS)
(e.g., MonaLisa [11]) and the local resource scheduler. Details on the kind
of information a local resource scheduler can supply are given in [1]. With
the detailed information gathered, the next step is to decide which resource
(or set of resources) will be used.

7. 3 . 3 Phase 3: Job Submission

Strictly speaking, job submission does not belong to resource allocation;
we briel y mention it just for completeness. Once resources are chosen, the
application can be submitted to the resources. Job submission may be as
easy as running a single command or as complicated as running a series
of scripts and may or may not include setup or staging.

7.4
In general, there are i ve main approaches to achieve efi cient grid resource
allocation: performance prediction, resource matching, economic sched-
uling, peer-to-peer method, and advanced reservation.

Major Approaches

7. 4 .1 Performance Prediction

Grid resource performance delivering to any given application l uctuates
dynamically in any time duration. A grid scheduler needs to predict what
the deliverable resource performance will be for the time period in which
a particular application component will eventually use the resource [12].
In order for a scheduler to determine which resource is most valuable in
terms of performance, there must be some kind of resource evaluation or
characterization methods for the scheduler to predict the performance of