online and strategic aspects of network resource management algorithms

Today’s communication networks provide a wide range of algorithmic challenges for managing network resources, such as memory buffers, computation power, bandwidth, as well as other resources. Challenges include developing algorithms, which are efficiently computable on one hand, and achieve optimal or near optimal optimization on the other hand. Additional aspects of algorithmic design in network environments include online computation, where decisions have to be taken before the entire input is known, and mechanism design, where the algorithm designer is aiming to achieve the cooperation of selfish agents, distributed over the network, who participate in the execution of the algorithm. In the first chapters of this dissertation we consider the task of online packet scheduling of a single FIFO queue. For non preemptive queues and arbitrary packet values, an online policy that nearly matches the lower bound for online scheduling is given. Randomized algorithms are considered for preemptive queues, including an algorithm for two packet values, which performs better that any deterministic algorithm. In the remaining chapters we consider mechanisms for allocation problems, such as machine scheduling and resource allocation. At first, resource allocation under budget constraints is considered. This is a special case of combinatorial auctions with compliment free valuations, as well as a relaxation of the general assignment problem [72, 24]. At first, we develop several non strategic algorithms for this problem. Following, we derive a general sufficient condition for designing incentive compatible mechanisms when the valuation functions of the participants are one dimensional. Using this, we present truthful algorithms for the auction with budget constraints problem, as well as for machine scheduling on uniformly related machines, when machines and job owners may be strategic. Preliminary results of this dissertation appeared in [7, 8, 6, 5, 9].