Accession Number : AD1030686


Title :   Efficient Orchestration of Data Centers Via Comprehensive and Application Aware Trade Off Exploration


Descriptive Note : Technical Report


Corporate Author : Naval Postgraduate School Monterey United States


Personal Author(s) : Bairley,Alan M


Full Text : https://apps.dtic.mil/dtic/tr/fulltext/u2/1030686.pdf


Report Date : 01 Dec 2016


Pagination or Media Count : 107


Abstract : Software-defined network (SDN) orchestration, the problem of integrating and deploying multiple network control functions (NCFs)while minimizing suboptimal network states that can result from competing NCF proposals, is a challenging open problem. In this work, we formulate SDN orchestration as a multiobjective optimization problem, present an evolutionary algorithm designed to explore the NCF tradeoff space comprehensively and avoid local optima, and propose a new application-aware approach that explicitly models resource preferences of individual application workloads. Further, we propose a new logical application workload (LAW) abstraction to enable precomputation of the required relative positioning of an applications virtual machines (VMs) and allocation of these VMs in a single atomic step, leading to online algorithms that are one order of magnitude faster than existing solutions for placing data center workloads. For an instance of the SDN orchestration problem subject to four independent NCFs attempting to optimize network survivability, bandwidth efficiency, power conservation, and computational contention, we demonstrate that our approach enumerates a wider range of, and potentially better, solutions than current orchestrators, for data centers with hundreds of switches, thousands of servers, and tens of thousands of VM slots.


Descriptors :   evolutionary algorithms , multiobjective optimization , genetic algorithms , heuristic methods , network architecture , throughput , energy consumption , network topology , virtual machines , computing system architectures , fault tolerance , workload , high performance computing , machine learning


Subject Categories : Operations Research
      Computer Systems


Distribution Statement : APPROVED FOR PUBLIC RELEASE