Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!rutgers!cs.utexas.edu!uunet!ibmpa!cslater
From: cslater@ibmpa.UUCP (Charlie Slater)
Newsgroups: comp.sys.ibm.pc.rt
Subject: IBM/4.3 4-port async driver
Keywords: 4.3 uucp asynchronous
Message-ID: <1259@ibmpa.UUCP>
Date: 8 Jun 89 17:44:43 GMT
Organization: IBM TCS Development, Palo Alto
Lines: 52


Status: IBM/4.3 4-port async problem

The driver for the 4-port buffered asynchronous communications card has
been modified.  With the new driver, uucp is stable at 9.6 Kbps and has
held the connection for as long as one hour at 19.2 Kbps.  Throughput
measurements for the transfer of a one-megabyte file between two RTs
connected via a null modem are as follows:


        uucp of a 1007616-byte file

Speed Setting             Time         Line Utilization
 9.6Kbps (typical)        1311 sec     80%
 9.6Kbps (best)           1183         89 
19.2Kbps (typical)         759         69 
19.2Kbps (best)            666         79 

The RT's used had 4 Megabytes of memory and no APC.  Note that these
performance measurements were done with the buffered 4-port card
(which contains 4 national 16550 UARTs each with a 16-byte FIFO).
The older, unbufferred 4-port card will not support uucp at speeds in
excess of 2400 bps.

Two characters per millisecond seems to be as much as the tested
configuration could handle.  For now at least, it would be a good idea
to plan on an aggregate limit of 19.2Kbps for simultaneous uucp
connections (2x9.6Kbps, 4x4.8Kbps, 8x2.4Kbps, or any other mixture
which totals to less than or eqaul to 19.2Kbps).

This driver has been used on "ibmsupt" to download netnews from uunet
using a Telebit Trailblazer modem with the uucp "g" protocol enabled.
Throughput measurements for batched net news files (~50Kbytes each) are
shown below:

Speed Setting             Line Utilization
 9.6Kbps (typical)        80%
 9.6Kbps (best)           89 
19.2Kbps (typical)        55
19.2Kbps (best)           67 

All of the kernel changes needed to implement the new driver have been
isolated to asynchronous communications support, and do not depend on,
or interfere with, other bug fixes.

The code is now being beta tested at an IBM Research location.  We
regret that it has taken so long to overcome the problem, and hope to
distribute the solution shortly.

charlie slater
uunet!ibmsupt!cslater
(415) 855-4413