🚀 go-pugleaf

I'm building an application using IPC under ESIX 5.3D (SYSVR3.2).  I
have a process which reads packets off an ARCnet card and shoots the
packets down different message queues depending on the packet type. 
A server process reads packets off the other end of each queue.  My
problem is that if one queue over-runs (I'm using the IPC_NOWAIT flag in
the msgsnd), all queues become clogged - a msgsnd to ANY queue returns
error EAGAIN.  The reader for the full queue never returns from the
msgget call.

I have a couple of questions

1.  Why does one queue filling stop all queues ?  My tunable parameters
are:  MSGMNB = 4096, MSGSSZ = 8, and MSGSEG = 1024.  If I understand how
these play together, this means the maximum number of bytes allowed on a
queue is 4096 (MSGMNB), and the maximum bytes pending system wide is
8192 (MSGSSZ*MSGSEG).  It looks like one queue filling up should leave
4096 bytes left for other queues.

2.  I use ftok(3C) with the same path parameter and different id
parameter to generate each message queue id.  Does this make the queues
"related" in some way ?

3.  What should I do when I detect a queue over-run ?  Should the writer
process read all the data out of the queue ?  Should the writer close
and re-open the queue ?

4.  As a side issue, would raising the tunable parameter MSGSSZ from 8
to 16 or 32 improve performance of system IPC ?  Has anyone played with
this value ?

Please email responses to ...!uunet!coplex!nos850!ezelmo!kenc.  I don't
think this system (kedeng.uucp) is in the maps yet.
-- 
Ken Cornetet                            Kimball Electronics
uunet!coplex!nos850!ezelmo!kenc         Phone: (812) 634-4482