Date: Sat, 14 Jul 2007 11:41:32 -0700 From: Brian Somers <brian@Awfulhak.org> To: Brett Glass <brett@lariat.net> Cc: freebsd-net@freebsd.org, Mike Tancsa <mike@sentex.net> Subject: Re: Bug in userland PPP LQR? Message-ID: <20070714114132.6b395616@dev.lan.Awfulhak.org> In-Reply-To: <200707120114.TAA28481@lariat.net> References: <200707110014.SAA02181@lariat.net> <ousa931n9u85mja8m4d26p0r3ol1g3h062@4ax.com> <200707120114.TAA28481@lariat.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, 11 Jul 2007 19:14:03 -0600 Brett Glass <brett@lariat.net> wrote: > At 06:23 PM 7/11/2007, Mike Tancsa wrote: > > >Did you try and use just LCP echo mode instead ? I have come across a > >number of devices (especially GPRS/EVDO cards) that seem to say yes to > >supporting LQR, but do not. Try instead lcp echo > > I will try it. (To be more specific, I am going to try > > disable lqr > allow lqr accept lqr > enable echo > echoperiod 12 set echoperiod 12 > so that the peer can get LQR if it requests it.) But since this would > just be working around the bug I think might be there, it would also > be a good idea to look at how the LQR counter is managed. From what > I can see, the problem is that the counter either is cumulative or > counts irrevocably up to 5 after one LQR packet is missed. The > reason why I'd like to see more eyes than my own on this is that > it's difficult to see how and where the LQR routines are invoked > and how they react to a pattern of missed and un-missed packets. I'd also add "set log +lqm" to your configuration. Disabling lqr should just remove the problem as there will be no calls to SendLqrReport(). Try adding the lqm logging with lqr enabled too. It'd be interesting to see what's actually being sent out. I expect unacknowledged LQR packets to be resent 5 times (exactly the same packet), and the 6th timeout to cause a line drop. The spec says that the peer may ignore an LQR request if it's under load, but that it must respond to a duplicate LQR request. My suspicion is that some implementations just ignore LQR altogether under load. These implementations should disable LQR if they can't implement it properly. Of course the *other* option is to implement an LQM strategy. I've never come up with anything that might really be useful though - except for suggesting that LQR is disabled. -- Brian Somers <brian@Awfulhak.org> Don't _EVER_ lose your sense of humour ! <brian@FreeBSD.org>
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20070714114132.6b395616>