Re: Dialup networking/ppp

[Ignatios Souvatzis <is%netbsd.org@localhost>]

On Tue, Sep 18, 2001 at 12:50:03AM +0100, Lars Hecking wrote:
>  
> >  There appears to be some improvement. I started a download via scp and
> >  a kernel compile, and the download does continue - mostly. Looking at the
> >  modem leds, it sometimes stalls for a few seconds, but then continues.
> 
>  Also, the rate of ser0 silo overflows has increased about tenfold,
>  compared to previous log entries:
> 
> Sep 18 00:34:19 localhost /netbsd: ser0: silo overflow
> Sep 18 00:35:00 localhost /netbsd: ser0: silo overflow
> Sep 18 00:36:42 localhost last message repeated 65 times
> Sep 18 00:46:50 localhost last message repeated 264 times
> 

Another improvement is possible (well, try it) by minimising the burst length.

in siop2.c: : line 615 of 1833, thats in the siopngreset() function,
it is currently set as:

        rp->siop_dmode = 0xc0;          /* XXX burst length */

this sets 16 longwords dma transfer,  5 clocks for the cpu.

Try to reduce this to 0x80, 0x40, or 0x00 (for 8:5, 4:5 or 2:5). Your
disk throughput will be worse, but the serial interupt function should have
less problems being fast enough.

I'm still not fully convinced (looking at the real numbers) that
there isn't a bad cable contribution to your problem. When the
transfer stalls, is the modem RTS light off, too?

Hm, btw.: for those 264 silo overflows, how much data did flow in? Look at
netstat -b -I ppp0; you should see an ibytes column. 329 sounds like a lot,
but some are inevitable with your hardware combination, and at 38400 bps
you can receive 2304000 bytes in 10 minutes. With a 7 bit deep hardware FIFO,
a bit over 0.01 % loss isn't too bad.

Regards,
        Ignatios