[4299] in linux-net channel archive
Re: bad PPP Performance (with and without IRQTUNE)
daemon@ATHENA.MIT.EDU (Edward Welbon)
Thu Sep 5 13:46:09 1996
Date: Wed, 4 Sep 1996 18:14:47 -0500 (CDT)
From: Edward Welbon <welbon@bga.com>
To: Simon <shimon@i-connect.net>,
Michael Neuffer <neuffer@goofy.zdv.uni-mainz.de>
cc: linux-kernel@vger.rutgers.edu, linux-net@vger.rutgers.edu
In-Reply-To: <Pine.LNX.3.91.960904124033.265b-100000@kyle.i-Connect.Net>
On Wed, 4 Sep 1996, Michael Neuffer wrote:
> Symptoms: system is normally lightning FAST. Compiles a complete kernel, on
> an Iomega Jaz drive in about 3-4 minutes (distclean, depend, make, zlilo,
> modules, modules_install).
Very good, I'd like to know more of the details of the configuration.
> PPP is on either Motorola BitSRFR PRO (115,200 on one B channel) or a new USR
> Sportster ``33.6 capable'', again, at 115,200.
>
> setserial /dev/cua{1,2} spd_vhi is done on both.
>
> Using CU, either on the built-in UARTs or the external one, serial I/O is FAST.
> Hard to measure but flies on the screen, with the ISDN linke visibly faster.
>
> Running PPP (pppd 2.2.0f) disconnects two/three times upon dialup from
> either line. Once connection is established, performance is horrendous.
> Typically ftp will do 800 bytes/sec on the ISDN line. Less on the
> modem.
Um, 800 bytes/second is a mite slow.
Is this MLPPP on two channels or is this one channel. I find that
non-PAP and non-CHAP connections are abysmal. On my BSPro (with 1H
firware) I get about 10KB/s with 115200, with 230400 baud, I get about
13.7Kb/s. This is the effective rate, I don't know what the bit rate
is (encapsulation and all that ya know).
I am running 2.2.0e and 2.2.0f (on the other system), the system that I
usually measure ISDN performance on is a 133Mhz 486. It has kernel
1.3.97, the kernel one the other system 1.99.9
It should work. I would be surprised to see that the P6 is the problem,
but I would also be surprised that the P6 is going to help get you over
13.7KB/s. There is encapsulation overhead everywhere. It will cost in
terms of bit bandwidth and processing time.
---
Ed Welbon; welbon@bga.com;
http://www.bga.com/~welbon/spider.html
