Budget-constrained workflow scheduler
Abstract
Scientific applications are very often described through workflows. We can easily find examples in all scientific domains: in astronomy (the Montage application, the RAMSES Universe simulator, etc.), in bioinformatics (Epigenomics, DNA sequencing, Proteomics, etc.), in geoinformatics (climate modeling, ocean current modeling, techtonic plate movement analysis, etc) and many others. The reasons for this are that some applications, as for example natural phenomenon simulators, have an inherent workflow-like structure while others are build by aggregating several smaller applications together which again results in a workflow-like structure.
With on-demand resources, the execution of an application now corresponds to budget.
The goal of this project is to implement a prototype platform that allows budget-constrained execution of workflow applications on an IaaS cloud.
This project will use Nimbus as the IaaS provider, DIET (with MADAG) as the workflow engine and the RAMSES adaptive mesh refinement (AMR) application as the test application.
Intellectual Merit
The current project will provide a prototype platform for the efficient use of on-demand resources in scientific workflow applications. The prototype platform is based on the state-of-the-art in on-demand resource provisioning and budget-constrained scheduling and is a practical application of the two.
Broader Impact
The biggest impact of the current project will be that is will allow scientists to outsource their experiments to IaaS Cloud platforms in a transparent way. Thus, research the scientific experimentation overheads are reduced which results in a more efficient use of the scientists' time.
Use of FutureGrid
FutureGrid will serve as the testbed for the prototype platform. It will serve as deployment environment for in development and especially experimentation phase for prototype validation.
Scale Of Use
The development phase, which will last one month, will need a small number of VMs (no more than 10) and a small number of barebone nodes (1 or 2).
The experimentation phase, which lasts another month, will need a large number of VMs (more than 10, variable by experiment) and a small number of barebone nodes (1 or 2).
Futuregrid is a resource provider for