Near-optimal dispatching policy for energy-aware server clusters

Abstract

A server cluster can be modeled as a set of parallel queues, and the dispatcher decides to which queue the arriving jobs are routed. We consider an energy-aware dispatching system in a Markovian setting, where each server, upon becoming empty, enters a sleep mode to save energy, and to activate the server after sleep incurs an additional setup delay cost. We seek to optimize the performance-energy trade-off by applying the so-called Whittle index approach. As our main result, we derive sufficient conditions for the system parameters under which the problem is provably indexable, and also determine the corresponding Whittle index values explicitly. Our numerical experiments demonstrate that the resulting energy-aware Whittle index policy is able to perform very close to the numerically solved optimal policy and outperforms all considered reference policies.

Characterization of the Gittins index for sequential multistage jobs

“…Another application area concerns dispatching (a.k.a. task assignment) problems in parallel queuing systems, where the state of the parallel queues is modified not only by the assigned jobs but also the completed ones, see, e.g., [16,8,15,3]. The application of the Whittle index approach is even more challenging when the state space of a bandit is multidimensional (as in our scheduling problem in this paper 7 ) since there is no natural order of indexes available but it has to be discovered while solving the problem.…”

“…The application of the Whittle index approach is even more challenging when the state space of a bandit is multidimensional (as in our scheduling problem in this paper 7 ) since there is no natural order of indexes available but it has to be discovered while solving the problem. In [4,5,3,6], we have, however, managed to derive the Whittle index for some problems where the state space of the bandit is multidimensional.…”

“…In our scheduling problem, bandits are represented by jobs. The state of a job is described by its attained service 3 a, which increases at rate 1 whenever served, until the service is completed. As presented in [13,12,1,2], the corresponding Gittins index G(a) (both for the open version and the closed version of the scheduling problem) can be characterized as follows:…”

High Availability in Server Clusters by Using Backpropagation Neural Network Method

“…The running server continuously will have problems. The problem might be occur are excessive traffic, request flooding, data center downs, hacker attacks, power outages, natural disasters, and single point of failure [2]- [4]. High Availability is the quality of service to continuously operate without failing for a long period even though there are disruptions threaten the available services [1].…”

“…Algorithms commonly used for High Availability on server computer clusters are round robin, static-rr, wight rr, lastconn, and source [2], [5]- [7]. This type of algorithm in the scope of the data center is classic algorithm.…”