Decentralized self-balancing systems

Abstract

The transition to Nano-scale devices is expected to open the way for highly complex parallel systems. However, decreased reliability and strict interconnect limitations are two important challenges that the devices need to overcome. To do so, such systems have to adapt to the changing workload and faults, employing decentralized protocols to coordinate among the many components. In this paper, we investigate algorithms for evenly distributing resources among locally connected components, so that the system can dynamically self-balance as the availability of resources changes. We propose two efficient Decentralized Protocols that achieve near-optimal resource distributions. The protocols are scalable and guarantee the desired fair distribution regardless of the interconnect topology.