Abstract: A computer system having a shared disk file system running on multiple computers each having their own instance of an operating system and being coupled for parallel data sharing access to files residing on network attached shared disks. Access to a file by a processor node is controlled by tokens transferred to the node from a token manager. To prevent another processor node from removing a token after the token has been received, but before it performs the operation on the file, each process can lock the token after it has been received. A node with a token can lock a byte range of a file, which byte range may include all or only some of byte range cornered by the token.

Abstract: In a fault-tolerant distributed file system, the server state needs to be reconstructed after the server restarts or when its function is taken over by another node. A crucial part of the server state is determining whether or not in-flight, directory-operations initiated by the clients have been completed. Described is a check-before-request scheme that solves this problem. A global lock is obtained on the directory or directories first. A check is made on the directories to see if the operation would succeed, and if so a request is made to the server for the operation. If the server were to fail during the execution of the operation, completion of the operation prior to failure can be determined by merely re-examining the directory contents.

Abstract: A system and method for allowing a distributed, coherent network filesystem to have virtual filesystem instantiations existent on any number of client nodes as well as on the server node that maintains the physical instantiation. On the server node only one image of the filesystem is visible even though two instantiations (physical and virtual) have been realized on that node. The rendering the physical instantation invisible and inaccessible is accomplished by performing a "takeover" of the physical filesystems data structures by the virtual filesystems mounting subroutines.

Abstract: A system and method for non-disruptive recovery from a file server failure. State information for a file system managed by a server is maintained among a plurality of client nodes of a distributed system. When a failure of the server is detected, all ongoing requests to server are suspended. Then, the state information in each client node is transferred to an identified back-up server. Based on the state information, the file server state prior to the failure is reconstructed. After the reconstructing the file system, the suspended requests are restarted on the back-up server.

Abstract: A system and method for avoiding serialization on updating the modification time of files in a cluster system. In accordance with the method, the modification time field of a file can be accessed by multiple readers or multiple writers and each client in the cluster system can update their own copy of the modification time. Whenever a client requests to read the modification time, the copies of the modification time are reconciled. The copies of the modification times are also reconciled when a cache flush or synchronization operation forces such reconciliation. The present system and method supports the requirement (of certain operating systems such as UNIX) that an explicit user-issued command to set the modification time is accomplished by granting an exclusive-write mode for the modification field.