From: hoffman@xdelta.zko.dec.nospam
Sent: Tuesday, June 06, 2000 10:20 AM
To: Info-VAX@Mvb.Saic.Com
Subject: Re: OpenVMS commentaries (was Re: Gartner commentary on
Wildfire)


In article <8hi52f$qbf@gap.cco.caltech.edu>, mathog@seqaxp.bio.caltech.edu (David Mathog) writes:
:...Actually, for many programs its already very simple. (Depending on the
:complexity of the Makefile and how well the code is written.)  The C RTL
:is much better than it used to be.  However the problems
:that keep biting me are relatively few, but also relatively hard to
:correct:
:
:  1.  select() works differently -> major rewrite

  To be fixed (COE and CRTL).

:  2.  fork() works differently -> rewrite

  To be fixed (COE and OpenVMS kernel).

:  3.  IO techniques may implicitly depend on file caching and/or
:      run like slugs on OpenVMS. -> much debugging to find problem
:      areas, possibility that no amount of tweaking will make performance
:      acceptably fast.  (Especially with programs that do a ton of
:      manipulation of small files.)

  To be fixed for read.  (XFC)   Full write caching within XFC is
  presently well along in development.

:  4. "records" are >64k, which makes RMS very unhappy. -> ifdef's all
:     over the place and dig, dig, dig for the sections of code which
:     trigger these problems.  They can be very hard to find, buried
:     in sections of code which don't execute until you're demonstrating
:     the program to somebody :-(.

  Donno, haven't looked at this particular aspect -- obviously this
  particular behaviour is tied in with the need for ir use of stream 
  I/O, and not particularly related to RMS record access.  (In this
  case, you really don't want RMS particularly involved at all.)

 --------------------------- pure personal opinion ---------------------------
   Hoff (Stephen) Hoffman   OpenVMS Engineering   hoffman#xdelta.zko.dec.com