Execution System Overview
The core activity amongst all of the execution system team members is
to develop, refine, and implement the notion of Virtual Grids (VGs).
This includes the VG abstraction itself, implementation technologies,
and underlying statistical resource classification techniques. To
effectively support applications while achieving efficient resource
selection and scheduling, VGs must enable applications to express
simply their desired resource abstraction and preferences. To develop
these abstractions, we have studied several leading grid application efforts,
abstracted their requirements, and evaluated the effectiveness of VG
abstractions for each. We then implemented the VG abstractions deemed
most useful in the Virtual Grid Execution System (vgES) and tested our
evaluations. This abstraction-implementation-evaluation process
continues today as we extend the vgES, notably the incorporation of
support for time (slotted VGs) and reliability (fault tolerance) in new
complex, adaptive workflow applications. We also continue to refine and
improve implementation techniques for core vgES tasks, such as finding
and binding (discovering and reserving) appropriate resources and
monitoring execution status.
More details about VGs can be found on the dedicated Virtual Grid page.
The overall architecture of vgES is given on the vgES Design page.
Other pages listed at left describe activities that directly support the development and evaluation of VGs. In particular, they discuss
- Defining VGs through the Virtual Grid Description Language (vgDL)
- Characterizing resources for use in VGs through NWS measurements and statistical inference
- Allocating VGs with appropriate attributes through the VGrADS Find-and-Bind mechanism
- Monitoring VG activity with the Mortar system
- Evaluating VG performance and scalability through grid modeling and simulation