Date: Wed, 20 Mar 1996 15:35:39 -0500 From: dennis@etinc.com (dennis) To: michael butler <imb@scgt.oz.au> Cc: hackers@freebsd.org Subject: Re: PPP bonding .. Message-ID: <199603202035.PAA05896@etinc.com>
next in thread | raw e-mail | index | archive | help
>Greg Lehey writes: > >> If anybody either side of the pond can give any input on channel bonding, >> I'd be grateful, and I'm sure Gary will be too. > >Speaking of bonding .. what's needed to allow two modem-based PPP interfaces >link to the same destination ? As far as I can tell .. > >i) you need one "virtual interface" at the route socket level with path >splitting underneath it. > >ii) the "splitter" needs to have an intimate knowledge of the transmit queue >lengths of each of the paths in order to choose the one with the least >latency. > >iii) Possibly, with assymetric or paths of different speed, some weighting >needs to be applied in addition to simple length assessment. > >What else needs to be addressed ? Can this be (realistically) done by >hacking on ijppp or is it a kernel-only job ? > >Linux seems to have this capability .. do we want (need ?) to be compatible >with them ? Nay...not really. BONDING is the wrong term here. MP (Multilink-PPP) is the standard, RFC 1717. We're doing it for sync...its not very difficult to do with a truly layered implementation. MP uses sequencing so that packets which arrive out of order are reassembled correctly..although you certainly don't want that to happen very often. Its not very good, but it IS the standard and if you're going to do something it ought to be MP. dennis ---------------------------------------------------------------------------- Emerging Technologies, Inc. http://www.etinc.com Synchronous Communications Cards and Routers For Discriminating Tastes. 56k to T1 and beyond. Frame Relay, PPP, HDLC, and X.25 for BSD/OS, FreeBSD and LINUX
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199603202035.PAA05896>