dc.contributor.authorZhou, Chi
dc.description.abstractCloud computing has become a popular computing platform for many scientific applications from various research fields. The workflow model is widely used by scientists to manage and analyze those large-scale scientific applications. Due to the pay-as-you-go pricing scheme, resource provisioning for scientific workflows in Infrastructure-as-a-service (IaaS) clouds is an important and complicated problem for cost and performance optimizations of workflows. The complexities come from severe cloud performance and price dynamics and various user requirements on performance and cost. IaaS cloud environment is dynamic, with performance dynamics caused by the interferences from concurrent executions and price dynamics like spot prices offered by Amazon EC2. However, existing studies of resource provisioning of workflows are not aware of the cloud dynamics, and assume static workflow execution time in the cloud. IaaS clouds usually offer different types of resources (e.g., virtual machines and storage) and workflow owners can have different optimization requirements on performance and cost. However, we find that most existing studies adopt ad hoc optimization strategies rather than a systematic approach. For different optimization problems, specific heuristics are designed for the specified goals and constraints and are not flexible enough for the various and evolving user requirements. To address the above issues, we propose three projects to achieve flexible and effective optimizations for resource provisioning of scientific workflows in IaaS clouds. Specifically, we propose a probabilistic scheduling system called Dyna to effectively optimize the monetary cost of workflows in the cloud. Dyna adopts a probabilistic QoS notion to explicitly expose the performance and cost dynamics of IaaS clouds to users. We develop an A⋆-based hybrid instance configuration method to reduce the expected monetary cost of workflows while satisfying user-specified probabilistic deadline guarantees. To simplify the complexities in cloud offerings and user requirements, we propose a transformation-based optimization framework named ToF. ToF abstracts the common performance and monetary cost optimizations as transformations and formulates six basic workflow transformation operations. An arbitrary performance and cost optimization process can be represented as a transformation plan (i.e., a sequence of basic transformation operations). All transformations form a huge optimization space. We further develop a cost model guided planner to efficiently find the optimized transformation for a predefined goal (e.g., minimizing the monetary cost with a given performance requirement). Based on the above two projects, we propose a declarative optimization engine called Deco, which considers both the cloud dynamics and flexibility of optimizations. Deco embraces a cloud- and workflow-specific declarative language for users to specify various workflow optimization problems. The declarative language is in support of the probabilistic QoS notion. We further propose a probabilistic optimization approach for evaluating the declarative optimization goals and constraints in the cloud. To accelerate the solution finding, we leverage the parallelism of GPUs to find the solution in a fast and timely manner. We integrate our systems into a popular workflow management system named Pegasus. Experimental results with real-world scientific workflow applications on Amazon EC2 and a cloud simulator demonstrate that (1) the cloud dynamics greatly affect the monetary cost and performance optimization results of scientific workflows; (2) our declarative language is expressive to describe a wide class of optimization problems for scientific workflows; (3) our systems are able to optimize the monetary cost and performance goals while satisfying probabilistic QoS constraints.en_US
dc.format.extent147 p.en_US
dc.subjectDRNTU::Engineering::Computer science and engineering::Computer systems organizationen_US
dc.titleSimplified and effective resource provisioning for scientific workflows in IaaS cloudsen_US
dc.contributor.supervisorHe Bingshengen_US
dc.contributor.schoolSchool of Computer Engineeringen_US
dc.description.degreeCOMPUTER SCIENCEen_US
dc.contributor.researchParallel and Distributed Computing Centreen_US

Files in this item


This item appears in the following Collection(s)

Show simple item record