Research On Network Virtualization And Monitoring Of Inter-Datacenter

With the dramatic increase in information service demand in recentyears, datacenter has been rapidly developed. Especially as emergence andwide spread of cloud computing technology, which changes the deliverymode of IT resource, decouples upper applications from substratehardware to better support provisioning of a variety of applications. Whilecloud computing brings all kinds of benefits to datacenter, it alsointroduces a lot of challenges to be addressed, such as network architecturedesign, multi-domain datacenter resource joint, datacenter monitoring,green datacenter. This article mainly aims at addressing the challenges ofmulti-domain datacenter resource joint and datacenter monitoring andpresenting solutions.Most previous researches about multi-domain datacenter resourcesjoint mainly focus on the static bulk data transportation betweendatacenters. However, except for bulk data transportation demandintroduced by data replication and synchronization, cloud datacenter alsohas the demand of adjusting application dynamically and sharing ofresources between multiple users, which needs isolation. This articleproposed a new solution which takes advantage of optical networkvirtualization to satisfy these demands. Optical network has a variety ofadvantages in implementing virtualization, such as high capacity,long-distance relay, high secrecy, application independency, low cost,multiple switching and transportation granularity, efficient managementand protection, abundant control interfaces. Based on analysis of computervirtualization, we implement network virtualization on management planeof optical network, since the management plane provides much more flexibility. This solution realizes virtualization characteristics includingabstraction, isolation, elasticity, programmability, authentication,authorization and accounting. Meanwhile, we show the viability of ourproposal with experimental demonstration and performance test over afield trial optical network testbed located in Shanghai. The resultsrepresent that this solution implements flexible optical networkvirtualization and provides basis for multi-domain datacenter resourcesjoint.Most of current researches and industry solutions about datacentermonitoring have specific system boundary, such as computing resourcesmonitoring or network resources monitoring. However, cloud computingmakes all kinds of resources couple more closely. For example,inappropriate migration of virtual machine may result in increasing loadfor network and vice versa. So datacenter needs integrated monitoringsolution to provide unified monitoring of all kinds of resources. Wepropose a new datacenter monitoring system to address these demands.This system uses sFlow to collecting and analyzing monitoring metricsdata, which realizes simple deployment, lightweight, high scalability andpush-based data collecting; Realize data storage similar to Round-Robindatabase based on MongoDB to satisfy all kinds of time-variation andgranularity needs for monitoring data, as well as the need for massmonitoring data storage; Use Node.js as data presentation to support lowclient requirements and rich data presentation. At the same time, wecarried out some testing, including test of sFlow agent cost, stress andsystem scalability. The results represent that this monitoring systemimplements light weighted, scalable and unified datacenter monitoring andprovides reliable and timely basis for datacenter management.