Browse Prior Art Database

Enabling Standard Caching for Magneto Optic File Systems

IP.com Disclosure Number: IPCOM000115528D
Original Publication Date: 1995-May-01
Included in the Prior Art Database: 2005-Mar-30
Document File: 4 page(s) / 176K

Publishing Venue

IBM

Related People

Dewey, DW: AUTHOR

Abstract

To improve the performance of Magneto Optic (MO) media, tailored file systems are often used. These file systems will track pre-erased space on the media so that data may be written with two-pass writes (write, verify) versus three-pass writes (erase, write, verify) generated by DASD file systems treating an MO device as a DASD. The presence of erased sectors, however, means standard caching code cannot be used since the erased sectors will generate errors when pages are fetched. Also the existence of two kinds of writes is not handled by caching code designed for magnetic devices.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 32% of the total text.

Enabling Standard Caching for Magneto Optic File Systems

      To improve the performance of Magneto Optic (MO) media,
tailored file systems are often used.  These file systems will track
pre-erased space on the media so that data may be written with
two-pass writes (write, verify) versus three-pass writes (erase,
write, verify) generated by DASD file systems treating an MO device
as a DASD.  The presence of erased sectors, however, means standard
caching code cannot be used since the erased sectors will generate
errors when pages are fetched.  Also the existence of two kinds of
writes is not handled by caching code designed for magnetic devices.

      To bypass these problems, MO file systems often communicate
directly with the device driver and bypass existing caching code.
They must then use their own cache strategy and code.  In the AIX*
system, this is referred to as raw I/O and although it solves the
problems, it means a MO file system cannot not use the Logical Volume
Manager or Buffer Cache of the AIX system.

      This article presents a method to allow a MO file system to
utilize a standard cache without having to modify the caching code.
When applied to AIX, this allows the file system to communicate to
the drive through the Logical Volume Manager and Buffer cache like a
normal DASD file system, but still track erased areas and generate
two and three pass writes.

The following problems are encountered when using a standard cache on
media with erased sectors:
  o  Read of an erased sector generates an error code from the drive.
      When a file system request issues a read of a sector the cache
      code will attempt to read a multi-sector page.  The page may
      begin in front of the needed sector and run beyond it.  The
      sectors around the needed sector may be erased and hence the
read
      of the page will fail.
  o  When writing a sector the cache will attempt to write a whole
      page out.  The file system is unaware of the other sectors in
the
      page and the write may cover erased sectors that the file
system
      will later treat as erased.  An MO file system needs greater
      control over what is actually written so that it may correctly
      issue two or three pass writes.  If a standard cache is used,
the
      actual writes cannot be controlled by the file system.

      One approach to this problem is to have the optical device
driver fill in zeros where erased sectors are encountered and
continue reading successive sectors until the full cache page has
been read in.  This can lead to excessive performance degradation.
Likewise, when writing the drive could issue two pass writes and if
the sectors were already written the drive would recover.  This also
leads to excessive performance degradation.

      The approach presented here is for the file system to
communicate to the driver the location of erased secto...