Grid Applications
Based on our earlier discussion, we can align Grid Computing applications to have
common needs, such as what is described in (but not limited to) the following items:
Application partitioning that involves breaking the problem into discrete pieces
Discovery and scheduling of tasks and workflow
Data communications distributing the problem data where and when it is
required
Provisioning and distributing application codes to specific system nodes
Results management assisting in the decision processes of the environment
Autonomic features such as self-configuration, self-optimization, self-recovery,
and self-management
Let us now explore some of these Grid applications and their usage patterns. We start
with schedulers, which form the core component in most of the computational grids.
Schedulers
Schedulers are types of applications responsible for the management of jobs, such as
allocating resources needed for any specific job, partitioning of jobs to schedule
parallel execution of tasks, data management, event correlation, and service-level
management capabilities. These schedulers then form a hierarchical structure, with
meta-schedulers that form the root and other lower level schedulers, while providing
specific scheduling capabilities that form the leaves. These schedulers may be
constructed with a local scheduler implementation approach for specific job
execution, or another meta-scheduler or a cluster scheduler for parallel
executions. Figure 1.2 shows this concept.
Figure 1.2. The scheduler hierarchy embodies local, meta-level, and cluster
schedulers.
The jobs submitted to Grid Computing schedulers are evaluated based on their
service-level requirements, and then allocated to the respective resources for
execution. This will involve complex workflow management and data movement
activities to occur on a regular basis. There are schedulers that must provide
capabilities for areas such as (but not limited to):
Advanced resource reservation
Service-level agreement validation and enforcement
Job and resource policy management and enforcement for best turnaround
times within the allowable budget constraints
Monitoring job executions and status
Rescheduling and corrective actions of partial failover situations
Later in this book, full treatment is provided for many of the most notable scheduler
and meta-scheduler implementations.
Resource Broker
The resource broker provides pairing services between the service requester and the
service provider. This pairing enables the selection of best available resources from
the service provider for the execution of a specific task. These resource brokers collect
information (e.g., resource availability, usage models, capabilities, and pricing
information) from the respective resources, and use this information source in the
pairing process.
Figure 1.3 illustrates the use of a resource broker for purposes of this discussion. This
particular resource broker provides feedback to the users on the available resources. In
general cases, the resource broker may select the suitable scheduler for the resource
execution task, and collaborate with the scheduler to execute the task(s).
Figure 1.3. The resource broker collects information from the respective
resources, and utilizes this information source in the pairing process.
The pairing process in a resource broker involves allocation and support functions
such as:
Allocating the appropriate resource or a combination of resources for the task
execution
Supporting users' deadline and budget constraints for scheduling optimizations
Load Balancing
The Grid Computing infrastructure load-balancing issues are concerned with the
traditional load-balancing distribution of workload among the resources in a Grid
Computing environment. This load-balancing feature must always be integrated into
any system in order to avoid processing delays and overcommitment of resources.
These kinds of applications can be built in connection with schedulers and resource
managers.
The workload can be pushed outbound to the resources, based on the availability state
and/or resources, and can then pull the jobs from the schedulers depending on their
availability. This level of load balancing involves partitioning of jobs, identifying the
resources, and queueing of the jobs. There are cases when resource reservations might
be required, as well as running multiple jobs in parallel.
Another feature that might be of interest for load balancing is support for failure
detection and management. These load distributors can redistribute the jobs to other
resources if needed.
Grid Portals
Grid portals are similar to Web portals, in the sense they provide uniform access to the
grid resources. For example, grid portals provide capabilities for Grid Computing
resource authentication, remote resource access, scheduling capabilities, and
monitoring status information. These kinds of portals help to alleviate the complexity
of task management through customizable and personalized graphical interfaces for
the users. This, in turn, alleviates the need for end users to have more domain
knowledge than on the specific details of grid resource management.
Some examples of these grid portal capabilities are noted in the following list:
Querying databases or LDAP servers for resource-specific information
File transfer facilities such as file upload, download, integration with custom
software, and so on
Manage job through job status feedbacks
Allocate the resources for the execution of specific tasks
Security management
Provide personalized solutions
In short, these grid portals help free end users from the complexity of job management
and resource allocation so they can concentrate more on their domain of expertise.
There are a number of standards and software development toolkits available to
develop custom portals. The emerging Web services and Web service portal standards
will play a more significant role in portal development.
