So far no problem has been  found during beta-testing. I have now removed
the old DISTRIBUTE processor and changed  the command entry point to be a
PASCAL routine.  The REXX code is  called as a subroutine  of this PASCAL
code to  handle delivery of files  and to forward jobs  to other servers.
For mail jobs (normally over 95% of the traffic), the REXX code is called
only if there are jobs to forward.  I haven't decided whether to keep the
list processor  support yet, so  I am  not converting the  job forwarding
code  to  PASCAL  at  this  point  (the  list  processor  part  is  using
custom-made interfaces which for now can  only be called from REXX, and I
don't want to waste time reworking them until I'm sure I want to keep the
list  processor  support).  The  reason  I have  doubts  about  the  list
processor is that JNET (and some older levels of RSCS) corrupt files with
multiple dataset  headers and Wingra is  apparently not going to  fix it,
since  it is  documented somewhere  that  such files  are not  supported.
LISTSERV can now cut  copies of the job faster than  RSCS would, since it
has the data in memory and the logic would be in PASCAL rather than REXX,
so there isn't anything to save at the system level; you're just shifting
some of  the load from  LISTSERV to RSCS. The  only reason for  using the
list processor  at this point  is to  decrease network traffic,  but that
only works  if the jobs don't  get corrupted. The decision  will probably
depend on what happens in the OHSTVMA  region when NCSU and UNC leave the
network, and on what happens to OHSTVMA itself.
 
An unrelated change to the domain resolution code has made it possible to
decrease the difference  between generated and relayed jobs  to an almost
negligible amount. Here are the latest performance figures:
 
+-----------------+--------------+---------------+
|    Job type     | Ratio (TCPU) | Absolute TCPU |
+-----------------+--------------+---------------+
| Small/relayed   |     4.57     | 0.658 & 0.144 |
| Small/generated |     4.74     | 0.687 & 0.145 |
+-----------------+--------------+---------------+
| Large/relayed   |     9.59     | 5.244 & 0.547 | (500 rcpts)
| Large/generated |    13.95     | 7.967 & 0.571 |
+-----------------+--------------+---------------+
| Huge/relayed    |    40.39     | 230.2 &   5.7 | (8500 rcpts)
| Huge/generated  |    45.90     | 312.1 &   6.8 |
+-----------------+--------------+---------------+
 
  Eric