Information Technology Reference
In-Depth Information
workflows at the workflow-level. Therefore, how to extend the current scientific
workflow temporal verification strategies to effectively monitor many parallel compu-
ting tasks is an open challenge.
3.3
Open Challenge #3: Temporal Violation Handling for Scientific Cloud
Workflows
Generally speaking, the objective for handling intermediate temporal violations in
scientific workflows is to reduce the response time of the subsequent workflow activi-
ties so as to compensate for the existing time deficit. In the conventional resource
environments such as cluster and grid, it is very expensive and time consuming to add
new resources such as virtual machines during workflow runtime. Therefore, we try
to utilize existing resources as much as possible with workflow rescheduling strate-
gies. However, without additional resources, some serious temporal violations may
not be able to be recovered. Meanwhile, due to the change of the local schedule, the
tasks of other workflow instances will also be affected and even violate their temporal
constraints. Therefore, there is obvious limitation in the conventional handling strate-
gies. The emergence of cloud computing can totally change such a situation. Current-
ly, in most public clouds, virtual machines can be easily provisioned in less than 30
seconds. In addition, many public clouds such as Amazon Web Service provide dis-
counted prices for reserved virtual machines and even bidding prices for spot
instances to further reduce the cost for dynamic provisioning. Therefore, it is now
becoming acceptable in both time and cost to add new resources at workflow runtime.
There are some literatures reporting the time for starting new virtual machines in
popular public clouds [38, 39]. The results for much more benchmarks on the perfor-
mance of computing, storage and network services in public clouds can be found at
[20]. However, for handling temporal violations, we need to not only consider the
time overheads for adding new resources but also the time overheads for data and task
re-allocation between the existing and new resources. It is necessary to estimate the
cost of the handling strategy and then make smart trade-off between violation han-
dling and the intermediate temporal violations. Therefore, it is still an open challenge
for the design of efficient and cost-effective violation handing strategies for scientific
cloud workflows.
4
SwinDeW-V Research Project
SwinDeW-V is an ongoing research project and part of the Swinburne Cloud Infra-
structure. SwinDeW-V focuses on temporal verification and serves as one of the key
functionalities in our SwinDeW-C cloud workflow system [34]. In this section, we
introduce the system implementation of the temporal verification framework in our
SwinDeW-C cloud workflow system.
SwinDeW-C (Swinburne Decentralised Workflow for Cloud) is located within
SwinGrid environment which contains many computing nodes distributed in different
places. Each grid node contains many computers including high performance PCs
