Virtualizing network file systems
Virtualizing network file systems(2009)
摘要
Over the years, centralized file server appliances based on standard protocols such as NFS and CIFS have been adopted as the de facto standard for file service. The scalability limitations of centralized file servers have led large organizations to deploy numerous independent appliances to meet huge increases in their storage demand caused by phenomenal data growth and more stringent regulatory compliance requirements. Unfortunately, this stop-gap approach brings administrators into another management quagmire: forcing them to manage numerous independent storage islands and fight bottlenecks and load imbalances at a high cost. In addition, this approach often causes significant outage in file service as well as other business critical services that depend upon file service, resulting in loss of revenue as well as direct and significant customer impact. This dissertation proposes a new virtual file service (vFS) layer to automatically manage file server infrastructure as a single resource pool with minimal or no attention from system administrators. vFS provides non-disruptive file service to clients even during server capacity expansion, server capacity shrinkage, server load balancing and other server maintenance events through protocol virtualization and transparent data migration. It consists of a dynamically growable set of light-weight vFS nodes placed between the clients and the file servers, and in achieving this, it requires no changes to the clients, the servers and the protocols they use for communication. vFS nodes serving the same virtual volume also form a federation amongst themselves such that they can be used interchangeably for load balancing and failover purposes. A proof-of-concept vFS prototype has been implemented on Linux with easily manageable software complexity. This prototype was used to measure the virtualization overhead for different NFS operations and the results indicate that most of the delay was caused by dual network traversals. A single vFS node without optimization is able to manage about 7 file servers in the experimental setup and adding more vFS nodes to a federation can aggregate the throughput of even more servers effectively. vFS nodes in the same federation are able to take over the responsibility of failed nodes quickly and transparently without clients noticing even for very strict client retry configurations. Transparent online data migration also has proved to be an effective mechanism in correcting load imbalance and access hotspots on file servers.
更多查看译文
关键词
light-weight vfs,new virtual file service,file server infrastructure,centralized file server,vfs node,virtualizing network file system,file service,load imbalance,proof-of-concept vfs prototype,file server,centralized file server appliance
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络