Re: LFB library

Jamal Hadi Salim <hadi@znyx.com> Tue, 01 July 2008 14:35 UTC

Message-Id: <TUE.1.JUL.2008.103557.0400.>
Date: Tue, 1 Jul 2008 10:35:57 -0400
From: Jamal Hadi Salim <hadi@znyx.com>
Organization: ZNYX Networks
Subject: Re: LFB library
Comments: To: Dong Ligang <donglg@mail.zjgsu.edu.cn>
Content-Type: text/plain
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit

Aron,

We've had support for at least 2-3 years S/FAPI for LFBs in our
implementation but contemplating getting rid of it altogether (it is
code maintainance overhead). There are many subtle practical issues that
are overlooked in the API unfortunately. It is still be useful to look
at that API as a reference point but keep in mind its
practicallity(sp?). It seemed that every time we had a new chip that
implemented differently we had to juggle the model.
I havent paid close attention to what Joel and Ligang had published, but
a quick litmus test would be to take 2 or more chip vendors of say an
IPV4 Forwarder LFB and make sure the model works for all.

We ended adopting (with some small caveats) the model defined by Linux
for the different LFBs that exist. Although slightly dated, if you look
at the different  "services" messages in
http://www.faqs.org/rfcs/rfc3549.html you'd get a feel for what we did.
On solaris or the BSDs interfaces like route sockets, pfkey etc will
probably give you a similar generic interface.
I was hoping to post the XML for the LFBs we implemented at some point.

cheers,
jamal

On Tue, 2008-01-07 at 10:11 +0800, Dong Ligang wrote:
> Dear Aron,
> My research group has provided more LFB definition. Please check enclosed
> file that have been submitted to ForCES group.
> Yes, NPF FAPI can be a good reference for defining LFBs.
> regards
> Ligang