On Tue, 5 Jul 1994 15:03:23 PDT Jim Smith said:
>I can address your first question:
>
>ADSM/6000 does not implement any special code to exploit logical volumes.
>Databases, log volumes and data volumes placed on JFS has out-performed
>logical volumes in tests.
>
>Jim Smith
>ADSM Technical Support

It would be interesting to see some test results comparing the two.  We have
tried it both ways, but didn't do any rigorous performance comparisons.  We
found no visually noticeable (i.e., not measured) performance difference.

I would have expected raw I/O (logical volumes) to perform better than JFS
files because I would think the code path lengths would be shorter.  This is
also the expectation documented on page 291 of "ADSM Presentation Guide" Red
Book (GG24-4146).  I quote from it:

  "With an AIX ADSM server you have the option of defining your database,
  logs, and storage pools as files in the journaled file system (JFS) and
  as files on logical volumes.

  "It is not clear at this time which option will provide better performance.
  We currently expect that using logical volumes will provide both better
  performance and better memory usage.

  "However, using JFS may not show a significant performance difference
  because AIX does a good job of buffering, especially if you have memory
  to spare.  The advantage of using JFS would then be ease of use, because
  it is a structured, hierarchical file system, with a directory and
  subdirectory structure."

Keep in mind that this was published back in December of 1993.

I suspect using JFS may do better when re-reading pages which are still
buffered (by JFS) in memory.  However, it's my understanding that you can
tune ADSM to use more memory for buffering also, which I would think would
have a similar beneficial effect, whether you're using JFS or raw logical
volumes.  Which buffering works better?  I dunno.

If you do use raw logical volumes, be sure you don't expand them after
ADSM has started using them.

..Paul