🚀 go-pugleaf

Hi, I'm involved in a project to connect a proprietary mainframe using
a Honeywell style cable, to a unix platform for the purposes of Unix
being a protocol gateway between the proprietary protocol and tcp/ip.
As a result, we are bypassing the network and session level netex
calls, and using only the netex driver level api (dread, dwrite, dconn,
ddisc, dwait). The mainframe side wrote their own driver to do those
as well (ccw based i/o, but proprietary).

Question #1: since we're not using transport level sequencing and
retransmission, we presume we need some ack/nak-ing in our protocol
levels that we add on top of driver layer netex. The DX will provide
a certain amount of buffering, but if a primed read is not available,
there is the provision to throw away the packet (under some hyper-
channel interfaces, particular hpux hpib, it would seem). Does any
one have a feel for the correct number of primed reads to have in
a typical data transfer transaction (please define typical from your
perspective).

Question #2: Netex seems primarily focused on transmitting large sized
packets (>4K blocks), Our gateway protocol requires the use of 4008
bytes per netex block, (or less). Does anyone have any experience with
the degradation factor of packet size over the hyperchannel.

In general, your comments about Hyperchannel operations using interfaces
with Unix would be highly appreciated. Please respond to gulati@
access.digex.com. Thanks,