Utility Driven Mobile-Agent Scheduling
Utility Driven Mobile-Agent Scheduling
Hi-index | 0.00 |
We consider a market of CPU time in a multiprocessor system providing the payable services. We suggest a market model under the following assumptions: (i) processors may be of different capacity; (ii) the processors' owners (suppliers) may have the different reservation prices; (iii) every user (consumer) has a linear utility function and one task within the considered period; (iv) a task can use different processors, but it cannot use two processors at the same moment. The main result is that all equilibrium CPU time distributions can be obtained as solutions to some linear programming problem, and equilibrium prices can be determined on the base of shadow prices for the same problem. An equilibrium distribution shows, what time each task should use each processor, but, generally speaking, it doesn't determine a scheduling. If the most labour-intensive job can be completed within the time-period to be distributed, using only the slowest processor, then the simple algorithm exists for constructing a scheduling corresponding to some equilibrium distribution. In general case such algorithm is unknown.