Study And Implementation Of High Performance Cluster File System In SAN Environment

This paper proposes a new cluster file system model ( SANFS ) for SAN based on the study of the main networking storage systems, and implements the model. The tests of the implementation verify the model's validity.SANFS can be of varied characters, such as high performance, high scalability, low cost and being easy to setup. It is applicable for such area as digital nonlinear video editing, science computing and VOD, all of which require both sequential I/O and high stability of I/O speed.Based on CIFS, SANFS can be implemented on most of the main operating systems nowadays. By utilizing the feature of SAN whose hosts can share the storage devices, the model enables the clients operate the storage devices directly under the control of MDS to improve the I/O throughput and the stability of I/O speed. The main ideas of SANFS include meta data's"cached dispersedly and controled centrally"and the file data's"distributed read-write". The main technic of SANFS vary from devices object cache, FDMT cache, meta data pre-allocate and pre-fetch, SANFS OpLock to SANFS DOOR Lock etc. The SANFS DOOR Lock model abstracted in the paper can be used for some other cachers, hence be of great value. Further, on the level of clients file system, SANFS presents support of MDS's LVM which improves the scalability.The paper give implementations of both SANFS's client on WINDOWS and SANFS's MDS on Linux, and presents analysis on the key technics of these implementations..Finally, this paper verify the model's validity and the technics'effect, by the tests on currently implemented version of SANFS.Finally, the paper tests the implementation. The test results are used to verify the SANFS model's validity and all of the technics'effect.