Subject: Re: More PPP questions
To: Bruce Albrecht <Bruce.Albrecht@seag.fingerhut.com>
From: John F. Woods <jfw@funhouse.com>
List: current-users
Date: 05/28/1996 22:21:28
> I've upgraded(?) from NetBSD-1.1 to NetBSD-1.2 alpha, running on an
> Amiga A3000 with 12 MB fast RAM, A2024 monitor, using the built-in
> serial port and a USR sportster 33K modem.  When I switched from
> NetBSD-1.1 to NetBSD1.2, my PPP performance was seriously degraded.
> I'm now losing packets (ping shows 5%-25% losses), rlogin sessions
> lock up for minutes at a time, I get VJERRS every couple of seconds
> and silo overflows every couple minutes.

Well, now this makes me suspect even more strongly that the lossage seen in
the 386 com.c driver has nothing to do with the com driver itself (hence the
calls for making it machine dependent seem more and more misguided).  However:

> My NetBSD-1.1 kernel had the lev6-defer and increased buffer (16K).
> My NetBSD-1.2 kernel has these and the two PPP options for VJ compression
> and deflate (I'd give the exact info, but I'm not at that machine).
> Was my 1.1 kernel really supporting VJ compression?

1.1 certainly had VJ header compression.  If you didn't explicitly disable it,
it should have been present.

> Does VJ compression require more bandwidth than my 25 MHz 68030 can provide?

Nah, VJ compression is real cheap.  However (speaking as the guy whose fault
PPP Deflate compression is), if you're actually *using* it on a 25MHz 68030,
you're likely sucking up fabulous amounts of CPU with it.  (My own tests showed
that at 56Kb and up it was pretty disappointing unless you had serious
processor power, unfortunately; on the other hand, that's more than twice the
line speed, and a 68030 wasn't one of the CPUs I tried.)  Tell syslog to
record local2.debug messages somewhere, and check to see whether pppd is
really negotiating the use of deflate compression; if it is, turn it off,
and see if that helps.