Re: [netmod] comments on draft-ietf-netmod-routing-cfg-04

[Ladislav Lhotka <[e-mail]>]

Hi Tom,

thank you for a detailed explanation. The moral of the story for me is that it is probably better to avoid FIB/RIB and use "routing table" everywhere, although it may look "less sophisticated".

As for the term "forwarding table", it is used in this sense in Junos documentation:

http://www.juniper.net/techpubs/en_US/junos12.1/topics/concept/routing-protocols-routing-databases-overview.html

Lada

"t.petch" <[e-mail]> writes:

> ----- Original Message -----
> From: "Ladislav Lhotka" <[e-mail]>
> To: "t.petch" <[e-mail]>; "Yi Yang (yiya)" <[e-mail]>
> Cc: <[e-mail]>
> Sent: Friday, August 03, 2012 12:10 AM
>> "t.petch" <[e-mail]> writes:
>> > <tp>
>> > In a sense I agree with you, in a sense I think the exact opposite.
>> >
>> > This topic of what tables there are in a box that routes packets has
>> > surfaced before.  I see, in a typical I-D produced about routing,
>> > reference to FIB and RIB, without any need to define or explain the
>> > terms, they are that well known, they are what routers and other
> boxes
>> > have as a matter or course.
>>
>> Yes, it's been my problem that I couldn't find a precise definition of
> these acronyms, and I understand most routing experts use the definition
> "you know it when you see it".
>>
>> >
>> > For me, it is the FIB that is always present; if you do not know
> where
>> > to send a packet, how to route it, then you are not an IP box of any
>> > shape or form - you must have a FIB..  In simple boxes, like Linux
> or
>> > Windows, then that is all you have.
>> >
>> > In more sophisticated boxes, 'proper' routers, for some meaning of
> the
>> > word proper, then you need much more data, which may closely model
> the
>> > FIB, as with RIP, or may be utterly different, as with OSPF, and for
>> > this, I see the term RIB used, initially with BGP but latterly with
> any
>> > routing protocol.  RIBs vary widely with the routing protocol and so
> are
>> > properly part of routing protocol extensions.
>>
>> So at the level of the present core routing model we are dealing with
> FIBs, but as soon as a BGP module augments routes with that
> sophisticated stuff a FIB gets promoted to a RIB, right? But this may be
> quite confusing from the point of view of the I-D text, and in this case
> I would probably prefer to avoid these acronyms entirely and use only
> "routing table".
>>
>> >
>> > So I think that this I-D gets the terminology quite wrong; as I say,
>> > this has surfaced before without any visible change to the I-D.  I
> was
>> > surprised that this topic attracted no comment from the Routing
>> > Directorate.
>>
>> Can you suggest what to do in order to get the terminology right?
>>
>> Anyway, I think that Yi Yang's comments were about "forwarding table",
> i.e. a (simplified) routing table which by definition contains only
> active routes.
>
> I think this a misuse of forwarding table:-(
>
> RIB is defined in RFC1771 (BGP) and later BGP RFC such as RFC2439 (Nov
> 1998) and RFC4771; it is clearly not the routing table, containing
> everything including the kitchen sink and its structure is specific to
> BGP.  But I see the term applied generically to what is learnt by an
> instance of a routing protocol, and so could be applied to the
> equivalent structures of EIGRP or IS-IS, which are very different in
> kind.  Sometimes I see it used generically to include everything learnt
> by all instances of all routing protocols but I think that that is
> wrong.
>
> FIB appears in RFC1812
> "The goal of the next-hop selection process is to examine the entries
>    in the router's Forwarding Information Base (FIB) and select the best
>    route (if there is one) for the packet from those available in the
>    FIB."
> which is what, in popular parlance, is a routing table.  For example,
> Moy, writing on OSPF, says
> "A router's routing table instructs the router how to forward  packets.
> Given a packet, the router performs a routing table lookup, using the
> packet's IP destination address as a key.  This lookup returns the best
> matching routing table entry ..."
>
> I would not expect the term FIB to include all the baggage that routing
> protocols provide nor for it to include any route which is not
> available, as far as the host or router knows, for immediate use.
>
> But the terms FIB and RIB are a bit specialised; I-Ds such as
> draft-uttaro-idr-bgp-persistence-01
> and
> draft-ietf-grow-simple-va-11
> use them without any explanation as do posters on many if not most of
> the WG lists in Routing and parts of Operations Areas, but that is
> probably not appropriate for this I-D, just as the terms are not used
> in, eg, RFC2096, which is entitled 'IP Forwarding Table MIB' but then
> defines 'IP CIDR Route Table', the DESCRIPTION of which is "This
> entity's IP Routing table."; oh dear, you can't win!
>
> So I think that conceptually, you should be defining the FIB, of usable
> routes from which the next hop or interface of a packet is selected on
> the basis of longest match, but probably calling it a route or routing
> table since that is the term most widely used.  Where individual router
> manufactuers do clever things on line cards, then that is best modelled
> by clever router manufacturers, even if they are all doing something
> similar - the speed and rate of change of those data structures probably
> makes them unsuitable for MIB, yang or anything else.
>
> You should allow for augmentation, or some such;  BGP revolves around
> peers and path attributes (RFC4273) and avoids all mention of FIB, RIB
> or the IP Route Table.  OSPF (RFC4750) covers interfaces, neighbors,
> areas and link states, and has a reference to the Forwarding Table.
> RIP(RFC1724) has interfaces and peers and refers to the  "the IP Route
> Database by RIP"
>
> Yup, you can't win, but referring to RFC1812 is unlikely to lose.
>
> Tom Petch
>
>> Thanks, Lada
>>
>> >
>> > Tom Petch
>> > </tp>
>> > Yi
>> >
>
>

-- 
Ladislav Lhotka, CZ.NIC Labs
PGP Key ID: E74E8C0C