Recovery schemes for high availability and high performance distributed real-time computing

Abstract

Clusters and distributed systems offer fault tolerance and high performance through load sharing, and are thus attractive in real-time applications. When all computers are up and running, we would like the load to be evenly distributed among the computers. When one or more computers fail the must be redistributed. The redistribution is determined by the recovery scheme. The recovery scheme should keep the load as evenly distributed as possible even when the most unfavorable combinations of computers break down, i.e. we want to optimize the worst-case behavior. In this paper we define recovery schemes, which are optimal for a number of important cases. We also show that the problem of finding optimal recovery schemes corresponds to the mathematical problem of finding sequences of integers with minimal sum and for which all sums of subsequences are unique.

Extended Golomb Rulers as the New Recovery Schemes in Distributed Dependable Computing

“…In this case the recovery schemes are implemented in the external systems as second, third, fourth,... alternative destinations (alternative cluster nodes) in case the primary, secondary,... destination goes down); • there is one network address for each computer and we use IP takeover (or similar techniques); • all the work performed by a computer is done by one process or a group of related processes that share local resources and thus must be moved as one unit. The results presented here can be easily generalized to the case with a number of independent processes on each computer using the same technique as in some of our previous papers [9,10].…”

“…Another algorithm, called Greedy, is presented in [9]. This algorithm generates the recovery schemes that give optimality for a larger number of cases than the Log algorithm, (i.e.…”

Improving fault-tolerance in intelligent video surveillance by monitoring, diagnosis and dynamic reconfiguration

“…In our system, however, we do not have the (non-volatile) memory capacity and the computational resources for a checkpointing approach. In general purpose distributed computing it is common to use redundant hardware and employ load sharing techniques to increase fault-tolerance (see, e.g., [5]). …”