From owner-freebsd-hackers  Sat Jun 28 06:56:21 1997
Return-Path: <owner-hackers>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.5/8.8.5) id GAA20298
          for hackers-outgoing; Sat, 28 Jun 1997 06:56:21 -0700 (PDT)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by hub.freebsd.org (8.8.5/8.8.5) with ESMTP id GAA20293
          for <hackers@FreeBSD.ORG>; Sat, 28 Jun 1997 06:56:17 -0700 (PDT)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.8.5/8.6.9) id XAA30041; Sat, 28 Jun 1997 23:53:27 +1000
Date: Sat, 28 Jun 1997 23:53:27 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199706281353.XAA30041@godzilla.zeta.org.au>
To: bde@zeta.org.au, Shimon@i-Connect.Net
Subject: Re: Clists limited to 1024 bytes?
Cc: bmcgover@cisco.com, hackers@FreeBSD.ORG
Sender: owner-hackers@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

>> You don't want to do this.  At 19200 bps, an 8K buffer takes more than
>> 4 seconds to fill.  A latency of 4 seconds is normally about 400 times
>> too large.  At such a low speed, there is normally nothing better than
>> delivering input in tiny bursts of 19-20 characters every 10 msec.
>
>Except with heavy PPP loads.  Right?

For ppp, it is better to deliver input in full packets a few microseconds
after each packet becomes complete.  The sio and cy drivers more or less
do this (they delivers input every 10 msec and a few microseconds after
a PPP_FLAG character is received).  To match the throughput of these
drivers, the hardware must support interrupting at least as often as
these drivers would deliver a packet.  The cy hardware has support for
interrupting on every PPP_FLAG (and some other) characters, but it is not
used by default because the extra interrupts cost more and the latency
was only reduced by a few hundred microseconds (at most the time to
fill the input fifo).  Perhaps this option should be used at low speeds.
The interrupt-on-special-character feature would be essential for devices
with larger fifos (unless the device supports ppp directly).

Anyway, 19200 bps is not a heavy load unless there are a lot of active
ports.  With 32 active 16550 ports it would be fairly heavy, but still
gives less than 6% of the throughput of a single 10Mb/s ethernet.

Bruce