Subject: Re: problems with ppp
To: None <email@example.com>
From: bryan collins <firstname.lastname@example.org>
Date: 09/28/1995 10:19:45
> >The remote end is a xylogics annex.
> What KIND of xylogics annex? The only one that has the proper hardware
> flow control (as far as I know, and I have admin'd a number of them
> from 1990-1994) is the Annex III.
> The portmaster has the proper flow control and can run up to 115kbps, the
> Annex III has proper flow control and can run at 38.4kbps, the lesser
> Annexen can do 38.4kbps, but have no (or limited to only 1 or 2 ports)
> flow control.
Its a Micro Annex, and it supports rts/cts to 57.6K, As I said in my original
post, I was running fine on linux, using the same hardware, (modem/annex etc) I
changed to NetBSD and it doesn't perform very well, seems obvious where the problem
> I don't recall you mentioning if you thought you were dropping packets or
> what. If you are on a lesser annex and see packet loss, I'd consider the
not the annex.
> I also don't recall if you mentioned using the i386 architecture or not,
> but if you are...
> Other observations from my use of PPP on NetBSD (and hearsay about the
> other *BSDs) is that they don't throttle the rts/cts lines at the driver
> level, they require ioctls from the higher levels to do it (which is IMHO
> really stupid, keep the IOCTL support, but when running in hard flow
> control mode, do it from the IRQ handler.) Regrettably (as implied by my
> last soapboxing) even if the drivers DID want to employ flow control at
> the lowest level, it would have to be done in software somehow, because
> the cruddy UARTs used in PCs (including the 16550Xs) have no hardware
> support for RTS/CTS flow control.
> You mentioned playing with 1.0 and -current...
> pppd in 1.0 had a bug that caused it to spin in a while loop somewhere
> where it continued to operate, but never slept. This resulted in adding a
> load of 1 to your system... and most probably slowing things down a
> little. There were patches for this bug MANY months ago that worked fine
> for me... this bug was in pppd, not the kernel, so it probably impacted
> all architectures. I know the patches for this went into -current because
> I have seen them there.
> Hope this helps.
I'll play around with it, will see how i go.
Coombs Computing Unit
Australian National University
Canberra, Australia, 0200
Tel: (06) 249 5586
Fax: (06) 257 1893