On Fast And Coordinated Data Backup In Geo-Distributed Optical Inter-Datacenter Networks

With the development of Internet Information Technology, entreprise or individ-ual’s dependence on information data and information systems has been increasing rapidly. Datacenters have been considered as the core part of the information and data for they can be stored centrally, thereby satisfying enough user needs. Due to cloud computing and a variety of new applications such as flow media, social networks, etc, the explosive growth in the number of users has come into being. Facing the expo-nential growth in data flow, datacenters acquire tens of thousands of servers and high bandwidth switches. However, electric switches need to consume large amounts of en-ergy to deal with the huge data demands. Optical Networks can be great alternatives because of their high throughput, low latency and low power consumption and have gained widely attention in recent years. Hence, the background network in this work focuses on optical inter-datacenter network.As datacenters are vulnerable to natural disasters such as flooding and earthquake, which may cause irreversible damage to both entreprises and users. In addition, the advent of these disasters are unpredictable. Therefore, preventive measures should be taken to prevent data loss causing by datacenter destruction. In order to prevent data loss, organizations usually take advantage of periodic backup mechanism. The rational of this mechanism is to create data redundancy distributed among different geograph-ical datacenters, that is to say, if a local datacenter is destroyed, data can be restored by the data from other datacenters. Periodic data backup indicates that a datacenter’s data is periodically backed up to another datacenter. However, for data transfers during backup process inevitably require large amounts of bandwidth resources, thus affecting other network services on the network. In order to reduce impacts on other network ser-vices, backup process should be scheduled when network traffic is relatively low and completed as soon as possible. Therefore, this work concentrates on how to improve the transmission efficiency of data backups among multiple datacenters.This work use optical interconnected datacenter network as the background and the network between datacenters is WDM optical network which can obtain optical-electric-optical transformation. Within the fiber in optical networks, signals are carried on by optical wavelengths. In addition, one-on-one backup model is used in our work where a datacenter can only choose a unique backup datacenter site, at the same time, it can only receive data from one other datacenter. In order to reduce the time impacted by backup processes on other network services, this work studies how to minimize the time needed for one-time backup, that is, minimize backup window. To achieve this goal, we consider coordinately optimizing the selection of backup sites and selection of backup paths.This work considers time-varying networks, that is, the parameter of time is sliced into equal-length time intervals. In each time interval, selection of backup sites and selection of backup paths are obtained in order to minimize total backup window as a whole. An integer linear programming model (ILP) model is designed under a series of constraints, thus determing the selection schemes of backup sites and paths for each time interval. However, ILP has rather high time complexity and is only suitable for small scale problem with less variations. In view of this issue, this work proposes an array of heuristic algorithms and theoretically analyses time complexities of all the algorithms. Through simulation, this work compares the result of different algorithms. In addition, via CPLEX optimization tool, a small scale problem of ILP is calculated whose results work as the benchmark of other heuristic algorithms.Slicing time parameter into different time intervals and re-optimizing within each time interval requires re-allocating network resources frequently, which bring a num-ber of operation overload to the network, that is, the complexity of network operations (operational complexity). On the other hand, if network is not optimized in time, cur-rent network status cannot be updated and thus network resources are not made full use of. This phenomena will cause network performance degradation. In response to this dilemma, we additionally propose several heuristic algorithms for balancing backup window and operational complexity. Simulations are made to compare the results of different algorithms and sum up which algorithm best balances the two parameters.