The Research Of Wide-Area Network Storage Virtualization

With the rapid development of applications, the complexity of large-scale storage systems grow fast as information resource exploding. It's more important that sharing information remotely then before. The trend is to interconnect the storage resources distributed in wide area network and use network storage virtualization to hide the details of those physical storage devices and map all kinds of hybrid storage devices into a single storage image. It can be transparent to users and provide a centralized storage management view only.It's worth study that how to coordinate so many kinds of hybrid storage devices and middlewares, how to gain high availability over instable underlay network, how to improve access speed over relative narrow bandwidth links, and how to keep some degree autonomy as resource sharing, which are researched in this thesis.First, we built a Storage Virtualization Volume Management System (SVVMS) as the platform of research based on hybrid underlay devices using asymmetrical virtualization. It extends volume management used in stand-alone machine and local area network into wide area network, and brings all kinds of storage network into the framework, such as SAN, NAS, JBOD, etc. It forms a virtual storage pool with single I/O addressing space for upper application. SVVMS has more adaptability than current volume manager because it uses Storage Resource Information Server (SRIS) and strict object owner mechanism.Second, after analyzing distributed network RAID in detail, we designed a Multi-Copies Striping Policy (MCSP) with hotspot migrate based on WAN. It places strip among all nodes in the storage networks through extended RAID 5, and uses multi-copies over instable WAN to gain expected availability. The copies migrate among the nodes based on access localization and threshold value, to gain the balance between efficiency enhanced by multi-copies and depression by consistent maintenance. Our results showed that MCSP works well, and the amount of copies is kept away from making more overhead while meeting expected availability.Then we studied the topology formed by wide area network storage virtualization, found out the shortage of existing simple hierarchical models, and designed Hierarchical Storage Network Topology (HSNT) to discovery topology and design fabric. In discovery process we brought forward IP path merge and trim algorithm fitting the features of storage network. It's a NP problem to design a fabric. We use Prufer code to generate random topology, then use heuristic genetic algorithm to gain optimize result. We also gave a simulation to simulate the automatic designing of fabric.Based on the above MCSP and HSNT, we extended cooperation cache into WAN and devised Hierarchical Unified Memory Cache (HUMC). It uses memory in remote computers unified as system global cache, then store strip copies as cache of local hard disk. It uses...