Richard Gambrell ([log in to unmask]) wrote:
: According to Isaac Blake:
: > their old DTC connections.  In testing we found a 19.2K DTC connection out
: > performed the NS/VT almost every time from a PC.
 
: That is interesting, we have found with 486 PCs that NS/VT outperforms
: a DTC  everytime for DOS based terminal emulators like Advancelink
: or ms92. We have been moving to WFWG 3.11 and win92 (using Microsoft's
: tcp-ip-32), we still get good performance, but it isn't nearly as
: fast as it was using Dos, with windows is it more like the DTCs.
 
This is interesting... and also tells me that I'm not insane (yet..)
I would expect (and I see) better performance through LAN using
WRQ stuff than I get from 19.2 kbps DTC port.  It makes sense since
LAN can get the stuff through much faster than a serial DTC port.
 
Maybe the case that Isaac refers to is a matter of running an application
that generates a big number of individual packets to paint the whole
screen instead of having a few fairly large writes.  I cannot see how a
DTC port could ever outperform a TCP connection over a LAN if the terminal
writes are few to paint the screen.
 
If the number of writes is great, than a VT-connection will suffer as it
requires almost twice the number of process dispatches - i.e. every terminal
write through DTS-io path can run on user's stack all the way out, a write
over VT runs on user's stack only for the time vt_ldm needs to process it,
then the vtserver needs to be dispatched to send the data out to the LAN...
Inbound path - I think - does not require any more process dipatches than
is needed for DTC attached terminal.
 
 
Eero - HP CSY Networking lab.