Citation Context

...rate logs for each node stored onsshared disk. Like GPFS, it uses a token-based distributed lock manager. A Frangipani file system resides onsa single, large (264 byte) virtual disk provided bysPetal =-=[19]-=-, which redirects I/O requests to a set of Petalsservers and handles physical storage allocation andsstriping. This layered architecture simplifies metadatas13s14 management in the file system to some...

Citation Context

...Ssthat allows multiple nodes to access data on sharedsdisks in an XFS file system. However, only one of thesnodes handles all metadata updates, like other SAN filessystems mentioned above.sFrangipani =-=[18]-=- is a shared-disk cluster file system thatsis similar in principle to GPFS. It is based on the same,ssymmetric architecture, and uses similar logging,slocking and recovery algorithms based on write-ah...

Citation Context

...lance data placement for throughput orsspace utilization.s2.2 Large Directory SupportsTo support efficient file name lookup in very largesdirectories (millions of files), GPFS uses extensibleshashing =-=[6]-=- to organize directory entries within asdirectory. For directories that occupy more than onesdisk block, the block containing the directory entry forsa particular name can be found by applying a hashs...

Citation Context

...e the file system is mounted orsevery time that one of the nodes in a cluster goes down.sInstead, GPFS records all metadata updates that affectsfile system consistency in a journal or write-ahead logs=-=[8]-=-. User data are not logged.sThere are two approaches to achieving the necessaryssynchronization:s1. Distributed Locking: every file system operationsacquires an appropriate read or write lock to synch...

Citation Context

...les across disks, involve a great amount of work.sGPFS performs its administrative functions in parallel as well.sGPFS achieves its extreme scalability through its shared-disk architecture (Figure 1) =-=[2]-=-. A GPFS system consists of the clustersnodes, on which the GPFS file system and the applications that use it run, connected to the disks or disk subsystems over asswitching fabric. All nodes in the c...

Citation Context

... for cluster computers that provides, as closely as possible, the behavior of a general-purpose POSIXsfile system running on a single machine. GPFS evolved from the Tiger Shark multimedia file system =-=[1]-=-. GPFS scales to theslargest clusters that have been built, and is used on six of the ten most powerful supercomputers in the world, including theslargest, ASCI White at Lawrence Livermore National La...

Citation Context

... listed above improve scalability bysoptimizing – or in some cases avoiding – the use ofsdistributed locking.s3.2 The GPFS Distributed Lock ManagersThe GPFS distributed lock manager, like many otherss=-=[2, 9]-=-, uses a centralized global lock manager runningson one of the nodes in the cluster, in conjunction withslocal lock managers in each file system node. Thesglobal lock manager coordinates locks between...

Citation Context

...ash value to use to locate thesdirectory block containing a given name. Hence aslookup always requires only a single directory blocksaccess, regardless of the size and structure of thesdirectory file =-=[7]-=-.s3 Managing Parallelism and Consistencysin a Clusters3.1 Distributed Locking vs. CentralizedsManagementsA cluster file system allows scaling I/O throughputsbeyond what a single node can achieve. To e...

Citation Context

...quential, as well as various forms of stridedsaccess patterns. For applications that do not fit one ofsthese patterns, GPFS provides an interface that allowsspassing prefetch hints to the file system =-=[13]-=-.sStriping works best when disks have equal size andsperformance. A non-uniform disk configurationsrequires a trade-off between throughput and spacesutilization: maximizing space utilization means pla...

Citation Context

...rchitecture to a storage area network (SAN)senvironment by allowing the file server clients to accesssdata directly from disk through the SAN. Examples ofssuch SAN file systems are IBM/Tivoli SANergy =-=[14]-=-,sand Veritas SANPoint Direct [15]. These file systemsscan provide efficient data access for large files, but,sunlike GPFS, all metadata updates are still handled by ascentralized metadata server, whi...