Algorithms For Survivability And Performance Optimization In Distributed Data Center

Data Center Network(DCN) is the supporting platform in cloud services. With the fast growth of cloud services, survivability and reliability issues in DCNs become increasingly important. The current data center architecture, which is based on electronic switches, centralized deployment and management and conventional multi-stage switch network, cannot satisfy the requirements of low-delay, low-cost, survivability, scalability, design flexibility, high-availability and for future cloud services. As optical switch offers high bandwidth, low latency and reduced energy consumption, it becomes a promising switching solution for future DCN. Based on all-optical switching and optical connections in the backbone network, we design the recovery of single core switch failure and service performance optimization of data centers separately.Because the core switches play an important role in both external and inter-node traffic switching, a single core switch failure may lead to traffic congestion or collapse of a DCN. Based on the scalable DCN architecture, in this paper we study two issues about survivability. The first one is fast protection of DCN traffics against a core switch failure while deploying the components of DCN. And the second one is how to deploy a backup core switch to fight against an arbitrary single core switch failure while minimizing the total protection cost. Upon the failure, all unaffected traffics keep their original routes without the need of reconfiguration, and the disrupted traffics are rerouted to the backup core switching capacities for fast traffic recovery. Under the survivability constraint and a given set of DCN external demands and internal traffic patterns, we formulate two integer linear programs(ILP). One is to distribute the DCN across a given optical network, such that all external demands can be satisfied and the DCN deployment cost can be minimized. And the other one is to place the backup core switch at a proper node and determine the fast traffic recovery mechanism. We validate the proposed design by extensive simulation studies.Generally, DCs are distributed at different nodes across a given optical network, and users access DCs through predefined routes. From the perspective of service providers, it is important to offer cloud service with high quality while taking service delay and service failure rate into consideration. In this paper, we study the service routing problem in a given optical network to provide cloud services with high quality, where failure probability and transmission delay of each link are considered as two coupled parameters. The objective is to minimize the total routing cost and design the access routes for the demands originated from each node. To combine together the corresponding delay and the failure probability of each link, we introduce failure delay cost factor which can be flexibly adjusted to control their relative importance(i.e., the weights). Based on mathematical approximation, a novel method is proposed to compute the failure probabilities of individual service paths. This translates our objective function into a linear expression. Then, we propose both an ILP and a heuristic to solve the problem. The proposed cloud service routing scheme is validated by numerical experiment.