PPP imbalance under 2.0.33

Scott M. Long (maxwell@europa.com)
Mon, 19 Jan 1998 14:25:27 -0800

I'm not sure if this has always been a problem, or has just appeared in
kernels 2.0.32 and 2.0.33 since I haven't really used anything else.

I've noticed that when I'm doing something like a big download (e.g.,
gcc- from sunsite) via PPP and I try to do anything else with the
link, like a DNS lookup or a telnet to my service provider, PPP seems to
starve the previous connection.

Example: I'm ftp'ing bison from sunsite.unc.edu, through Lynx. It's
coming in at about 3.0kbps, which is a fair rate for me. Now I go to
another VC and telnet to my service provider to check my mail. The
download freezes, and it appears as if all the bandwidth on the PPP link
is devoted to the new telnet connection. But the telnet is sluggish
(takes 10 seconds to update a screenful of info). I painfully read my
mail, and close the telnet. Now I have to wait about 5 minutes for the
ftp connection to come back to life -- 5 minutes before I see ANY
activity at all. And then I have to wait another 10 minutes or so for
the 3.0kbps rate to come back.

I don't know much about PPP, so this may be a symptom of the protocol,
but I know I don't have this problem under Windows 95. In fact, I
usually have about 10 downloads going under Netscape at once, so that I
can take a shower or something while they complete. And they all seem to
get their share of the bandwidth.

I figure this problem could be occuring at any one of these three places
(but I may be wrong):

1) The ppp driver itself (ppp.o) is doing some weird stuff.

2) pppd, whatever role it plays, is being unfair somehow.

3) The socket layer is broken in some way. But this would affect my
local ethernet segment, and I haven't seen anything like that happening.

What's up? I hate doing all my downloads under Windows just so that I
have a non-sluggish connection.