IBM recommends to use NET VPs for onsoctcp/ontlitcp and CPU VPs for
onipcshm. So does Art.
But what about onipcstr on Unix?

The recommendation is not platform specific, so the use of NET VPs for
network connections is probably true for Unix and Windows.
What about onipcnmp on Windows?

From what I read, a NET VP for TCP/IP can handle up to 300 connections
performant.
For shared memory connects the advice was to configure as many as CPU
processors.
Depending on the answers to the questions above, how many processors for
the pipe connections should be configured?

Regards, Frank

On 28 Apr, 21:09, Frank Langelage <fr... (AT) lafr (DOT) de> wrote:
I assume the OS calls for this onipcstr/onipcnmp can block hence use
NET VPs.

However if you have a lot of data transfers back to the client and
hence blocking will be of short duration and the overhead of a context
switch back to the cpu vp will be
significant then use CPU VPs.