Browse Prior Art Database

Serialization Mechanism for Parallel Journaling File System on a Cluster

IP.com Disclosure Number: IPCOM000112939D
Original Publication Date: 1994-Jun-01
Included in the Prior Art Database: 2005-Mar-27
Document File: 2 page(s) / 52K

Publishing Venue

IBM

Related People

Devarakonda, M: AUTHOR [+2]

Abstract

Disclosed is a method of serialization of accesses to file data in Parallel JFS (Journaling File System), a research prototype that runs on a cluster of RISC System/6000's*. JFS is the AIX* file system for a standalone RISC System/6000. In Parallel JFS, multiple instances of JFS, each running on separate AIX nodes provide shared access by directly reading and writing disk blocks representing file data and meta-data. The shared access provided by Parallel JFS is in the form of vnode and VFS operations.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 56% of the total text.

Serialization Mechanism for Parallel Journaling File System on a
Cluster

      Disclosed is a method of serialization of accesses to file data
in Parallel JFS (Journaling File System), a research prototype that
runs on a cluster of RISC System/6000's*.  JFS is the AIX* file
system for a standalone RISC System/6000.  In Parallel JFS, multiple
instances of JFS, each running on separate AIX nodes provide shared
access by directly reading and writing disk blocks representing file
data and meta-data.  The shared access provided by Parallel JFS is in
the form of vnode and VFS operations.

      Parallel JFS requires serialization of accesses to file data
and meta-data in order to preserve the integrity, consistency, and
support for POSIX semantics.  The disclosed method takes advantage of
the fact that the accesses are in the form of vnode and VFS
operations.  Each vnode and VFS operation is intercepted, and at the
point of interception, a cluster-wide lock request is made for each
file involved in the operation.  A read or write lock is requested
depending on the intended use.  The calling process is blocked until
the lock can be granted.  The locking scheme employs
multiple-readers/single-writer policy.  Upon acquiring the necessary
lock, the operation proceeds to work on the file(s) as in the base
JFS.  At the end of the operation, the locks are released.

      The cluster-wide lock and unlock primitives are implemented
using a token-based scheme.  When...