Information Technology Reference
In-Depth Information
Fig. 1.
A Generic Temporal Verification Framework for Scientific Cloud Workflows
The overall deadline is one of the most important QoS constraints for scientific
cloud workflows. In addition, for recurrent workflow applications (namely those
scientific workflow applications need to be repeated regularly), a SLA (Service Level
Agreement) for on-time completion is also required to establish different levels of
service quality (with different service prices) such as 90%, 99%, 99.9% and so on,
similar to the SLA for cloud storage and computing services [1]. Furthermore, since a
large-scale scientific cloud workflow may have a large number of activities and with
complicate process structures, local temporal constraints for workflow segments or
workflow activities are required to serve as the local milestones for monitoring pur-
pose. Therefore, after the setting of the overall deadline and its SLA, the next step is
the assignment of local temporal constraints.
Although temporal constraint setting only takes up a small effort in the temporal
verification framework but it is clearly that the quality of the assigned temporal con-
straints will have significant effect on the subsequent steps. For example, if the over-
all deadline is too tight, the chance that the targeted SLA can be met will be very
small regardless the subsequent monitoring and control strategies. In contrast, if the
assigned milestones are too flexible, the effectiveness of workflow monitoring will be
significantly deteriorated. Therefore, temporal constraint setting needs to consider the
trade-off between user requirements (e.g. workflow time and budget) and the system
performance (e.g. service response time and price). Meanwhile, the overall deadline
and the local temporal constraints need to be consistent so that the final on-time com-
pletion can be achieved with the progressively success of local milestones.
Generally speaking, there are two basic ways to assign QoS constraints, one is ac-
tivity-level assignment and the other is workflow-level assignment. Since the whole
workflow process is composed of all individual activities, an overall workflow-level
constraint can be obtained by the composition of activity-level constraints. On the
contrary, activity-level constraints can also be assigned by the decomposition of
workflow-level constraints [59]. The issue of automatically translating the overall
deadline into deadlines for the individual subtasks is investigated in [8, 21]. In [22],
based on a probabilistic temporal consistency model, a workflow temporal constraint
