China Grid and Related Dependability Research - Grid Computing: Infrastructure, Service, and Applications - page 106

level [70]. Our failure-handling framework supports checkpoint-
enabled tasks so that when task fails, the task can be restarted from
the recently checkpointed state. Beside the three levels of check-
pointing, we implement a workl ow-level checkpointing, which
dei nes the task as a workl ow and saves states of the atom-task as
checkpoints.
Workl ow: This is another technique to handle failures in grids. •

There are several workl ow methods for failure handling, such as
alternative task, workl ow redundancy, and user-dei ned excep-
tion handling. The main idea of this method is to construct a new
workl ow for task execution. The workl ow method can be regarded
as the composition of other failure-handling techniques.

4.5.4.3 Model of Policy Making

In this section, some basic dei nitions about failure-recovery policies are
presented, and some quality criteria for failure recovery are given. The
policy-making model is illustrated later.

4.5.4.3.1
We present some basic dei nitions about the failure-recovery policies.

Basic Dei nitions

1. Execution time ( Q et ): This is the execution duration of a task with
some failure-recovery methods. The task can be a single running
service or a composition of services. Execution time is important
for a failure-recovery method, especially in some mission-critical
applications. The execution time of the same task varies violently
with different failure-recovery methods.

2. Failure-recovery policy ( P ): This is a policy specii ed by the user
or t he s y st e m to ide nt i f y wh ic h fa i lu r e -r e cove r y t e c h n iq ue s s hou ld
be applied for a particular application. In this chapter, the failure-
recovery policy specii es the techniques presented above or speci-
i es the composition of them, such as replication with checkpointing
and replication with retrying.

3. Execution cost ( Q co ): This is the amount of money for a service
request to pay for executing the specii c operation for a service.
The execution cost can be computed using Equation 4.15, where
Q price is the price of the specii c resource, and R num is the number of
resources.

QQ = ¥ QR ¥ (4.15)

co price et num

4. Task: This is a state chart with a sequence of states [ t 1 , t 2 , . . . , t n ],
such that t 1 is the initial state and t n is the i nal state.