| Summary: | Grid Computing is a distributed computing paradigm where many computers often formed from different organisations work together so that their computing power may be aggregated. Grids are often heterogeneous and resources vary significantly in CPU power, available RAM, disk space, OS, architecture and installed software etc. Added to this lack of uniformity is that best effort services are usually offered, as opposed to services that offer guarantees upon completion time via the use of Service Level Agreements (SLAs). The lack of guarantees means the uptake of Grids is stifled. The challenge tackled here is to add such guarantees, thus ensuring users are more willing to use the Grid given an obvious reluctance to pay or contribute, if the quality of the services returned lacks any guarantees. Grids resources are also finite in nature, hence priorities need establishing in order to best meet any guarantees placed upon the limited resources available. An economic approach is hence adopted to ensure end users reveal their true priorities for jobs, whilst also adding incentive for provisioning services, via a service charge. An economically oriented model is therefore proposed that provides SLAs with bicriteria constraints upon time and cost. This model is tested via discrete event simulation and a simulator is presented that is capable of testing the model. An architecture is then established that was developed to utilise the economic model for negotiating SLAs. Finally experimentation is reported upon from the use of the software developed when it was deployed upon a testbed, including admission control and steering of jobs within the Grid. Results are presented that show the interactions and relationship between the time and cost constraints within the model, including transitions between the dominance of one constraint over the other and other things such as the effects of rescheduling upon the market.
|
|---|
