Load balancing in the presence of random node failure and recovery

Abstract

In many distributed computing systems that are prone to either induced or spontaneous node failures, the number of available computing resources is dynamically changing in a random fashion. A load-balancing (LB) policy for such systems should therefore be robust, in terms of workload re-allocation and effectiveness in task completion, with respect to the random absence and re-emergence of nodes as well as random delays in the transfer of workloads among nodes. In this paper two LB policies for such computing environments are presented: The first policy takes an initial LB action to preemptively counteract the consequences of random failure and recovery of nodes. The second policy compensates for the occurrence of node failure dynamically by transferring loads only at the actual failure instants. A probabilistic model, based on the concept of regenerative processes, is presented to assess the overall performance of the system under these policies. Optimal performance of both policies is evaluated using analytical, experimental and simulation-based results. The interplay between node-failure/recovery rates and the mean load-transfer delay are highlighted.

Probabilistic resource allocation in heterogeneous distributed systems with random failures

“…However, the available literature on distributed computing in such uncertain environments primarily considers reactive techniques, where a node failure is addressed only after its occurrence [7]. One of the few exceptions is the paper of Dhakal et al [12] that presents two preemptive load-balancing policies for a heterogeneous distributed computing system with wireless links between nodes. Preemptiveness in this case implies adjusting actions to compensate for the possibility of node failure/recovery.…”

Maximizing Service Reliability in Distributed Computing Systems with Random Node Failures: Theory and Implementation

“…Regarding the transfer times, our assumptions are justified according to our prior work [9], [10], [16] and the empirical data obtained from the experiments conducted over the DC architecture to be discussed in Section 3. In addition, we have assumed that the mean transfer time of the ith group of tasks being transferred to the kth node follows the first-order approximation:…”

“…In general, however, the above partitions p jk may not be effective and must be adjusted in order to compensate for the effects of the random transfer times. The load to be migrated from the jth to the kth must be adjusted according to what is called the load-balancing gain [9], [10], [16], [22], which is denoted as K jk , yielding…”

“…The role of LB in improving the performance of DCSs has been studied vastly considering a number of performance metrics; these include the average response time of an entire workload [1], [9], the probability of successfully serving an entire workload [10]- [16], the probability of serving a workload within a given amount of time [17], the average queue-length of a node [18], [19], and the total sum of communication and service times [5], [20]. In addition, the problem of LB has been studied under both static and dynamic scenarios.…”

“…In these works, the authors have considered random communication delays as well as random server failure. Additionally, in an earlier work we have studied the effect of node failure and recovery on the average response time of a workload served by a two-node DCS [10].…”