Resource Allocation for Distributed Systems with Fault Tolerant Nodes

Abstract

Resource allocation problems for distributed systems have been extensively studied for years. However, very few of the studies consider the failure/repair of the systems and the fault tolerant overheads. As the result, the solutions may not be applicable to fault-tolerant systems. In this paper, we study the resource allocation for a distributed system with fault tolerant nodes in which each node consists of several servers. One of the servers is the primary which serves user requests and the rest are backup servers. To maintain the consistency of the system, the primary server periodically checkpoints its status onto all backups. To study the resource allocation for such system, an approximate analysis is developed. With the model, an efficient allocation algorithm is presented which takes into account the failure rates and repair rates of the system and the fault tolerant overheads. From many experiments, it is shown that the algorithm gives the optimal allocations. The total system throughputs estimated by the algorithm are very close to the ones obtained by simulation. The algorithm, which incurs very little overhead, improves the system performance significantly over an intuitive allocation algorithm.