Browse Prior Art Database

Piecewise Directory Mapping for Write-Once Read-Many and Log Structure File Systems

IP.com Disclosure Number: IPCOM000114594D
Original Publication Date: 1995-Jan-01
Included in the Prior Art Database: 2005-Mar-29
Document File: 2 page(s) / 84K

Publishing Venue

IBM

Related People

Dewey, DW: AUTHOR

Abstract

In the IBM* High Performance Optical File System*, a hash table is used for the directory structure on Write-Once Read-Many (WORM) media. Each hash class contains a continuation chain comprised of directory entries. To enable enhanced write performance in certain optical library environments, a single hash class is used and a database of the locations for existing files is maintained. This allows the directory entries for multiple files to be written with a single disk access. However, when creating a file the file system must still determine if the file already exists. An efficient method of determining this must be used. A direct search of the single hash class on the disks is extremely slow. To avoid the expensive disk search the file system maintains a mapping of the directory.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Piecewise Directory Mapping for Write-Once Read-Many and Log Structure
File Systems

      In the IBM* High Performance Optical File System*, a hash table
is used for the directory structure on Write-Once Read-Many (WORM)
media.  Each hash class contains a continuation chain comprised of
directory entries.  To enable enhanced write performance in certain
optical library environments, a single hash class is used and a
database of the locations for existing files is maintained.  This
allows the directory entries for multiple files to be written with a
single disk access.  However, when creating a file the file system
must still determine if the file already exists.  An efficient method
of determining this must be used.  A direct search of the single hash
class on the disks is extremely slow.  To avoid the expensive disk
search the file system maintains a mapping of the directory.  This
map contains a hash value for every file in the structure and a
sector address containing the directory entry of the file.  To search
for a file, the map is examined and all entries with the same hash
value are read from the disk and compared.  There are several
problems with this approach:
  o  The entire map must be complete and in memory to be used.  This
      stresses memory resources in a library environment.
  o  The time to build the map can be extreme; hence, it is stored on
      the hard disk of the library controller.  However, the large
      capacity of the library means that not all maps can be
maintained
      on the hard disk but only a small percentage of them.
  o  The method cannot be used outside of a library.
  o  The disk cannot be gracefully interchanged to another library or
      work station without preparation or extreme access times.

      As library capacity increases with new designs, all these
problems are exasperated and the approach becomes less feasible.
This article presents a method of storing and updating this mapping
on the WORM disk itself to address the above problems.

      The structure of the hash class on the WORM disk is a
continuation chain.  This is a linked list snaked through the disk
where each node is a contiguous segment of sectors referred to as a
Directory Storage Segment (DSS).  The last DSS in the list contains a
blank s...