Version 2 (modified by 16 years ago) (diff) | ,
---|
Tendrils to the Control Framework WG
According to the GENI Design Principles (GDD-06-08 PDF), GENI must enable construction of slices that "arrange components into a representative network" in a way that offers "predictable and repeatable behavior".
What requirements does this place on the component interface?
- Slice construction. Is coallocation required? That is, is it ever necessary for an experimenter to allocate a defined group of resources as an atomic unit? Is it sufficient to allow experimenters to select from whatever resources are available at any given time? Or will some experiments require some set of resources matching a defined profile whenever the set becomes available?
- Slice fidelity. What level of assurance is required for the resources allocated to each sliver? Of course a Component Manager cannot guarantee that a component will continue to function at a given level. Does the slice service need it to provide any assurances at all, e.g., for "exclusive access to some subset of resources" in GDD-06-08? What assurances are needed to "support strong isolation between slices so that experiments do not interfere with each other", or to "provide enough feedback about what resources a slice actually receives to enable researchers to evaluate the validity of their results"?
- Policy control. According to GDD-06-08, GENI will be "sustainable" in the sense that it will provide incentives to contribute resources. For example, contributing resource providers might receive in return for "access to the resources of GENI as a whole". The GENI Principles also suggest that the GENI Science Board will "establish policies about how resources are allocated to research projects." What component interfaces and powers are needed to support these policy control services?