Re: [trill] Some question about TRILL IS-IS and Distribution Tree

[Donald Eastlake <[e-mail]>]

See below:

On Wed, Apr 25, 2012 at 7:01 AM, Nguyen Huu Tho <[e-mail]> wrote:
> Dear all,
>
> I have some questions about trill IS-IS and Distribution Tree. Anybody can
> help me ?
>
> 1. When TRILL use IS-IS, TRILL has a new type of hello message which is
> called TRILL HELLO.
> So, IS-IS HELLO of original IS-IS still have used or TRILL hello message
> includes all function of original IS-IS hello and original IS-IS HELLO isn't
> exist.

The TRILL Hello replaces the original IS-IS Hello on LAN links. It's
really just a variation and uses the same PDU type number and same
fixed PDU header. There is just some variation in its contents. RFC
6327 might be the best place to start concerning the TRILL Hello.

> 2. Is there any mechanism for each RBridge to choose the used distribution
> tree and advertise them to other RBridges

If there is only one distribution tree, obviously every RBridge has to
use it. If there is more than one distribution tree, how each RBridge
decides which trees it is going to use in general and which tree it is
going to use for any particular flow are out of scope of all current
TRILL documents.

RBridges can announce the distribution trees they might use with the
Trees Used Identifiers Sub-TLV. See RFC 6326.

> 3. In RFC 6325, In distribution tree part, When the RB having the highest
> tree root priority send a list of s tree roots nickname. Why we have the
> case s < k trees which all RBs in the campus must calculate ?

I don't quite understand your question. I assume you are talking about
Section 4.5 of RFC 6325. Not there are some change to this section in
draft-ietf-trill-clear-correct which has been judged to have WG
consensus.

The distribution tree determination logic is a bit complex because
different people wanted different things. Some thought that priority
should be adequate, with the k highest priority being computed. Others
wanted to always give an explicit list of the roots. The result is
somewhat of a hybrid which can express a variety of policies for
distribution tree determination.

The RBridge holding the highest priority nickname to be a tree root
can specify k, the number of trees they want computed, and can
separately specify a list of desired tree root nicknames of length s.
So, obviously, it can specify s < k. For example, k could be 3 and the
explicit list of root nicknames could be null in order to use the
three highest priority roots...

> 4. When Ingress RB have multi nicknames, there is any mechanism to choose
> which nicknames the RB will put in the Ingress nickname of TRILL header of
> this frame

Not currently. It is up to the RBridge.

> 5. When a Ingress RB1 receive a multi-destination frames, when the Ingress
> RB1 choose the distribution tree (have tree root is RB4). When the frame
> reaches to the RB4, whether the tree root send the frame to RB2 with
> outer.MacDA is All-RBridges or unicast Outer.MacDA ?

With the TRILL base protocol standard RFC 6325 any multi-destination
TRILL data frame being send on an Ethernet link is always sent with
the outer.MacDA All-RBridges. It does not matter where in the
distribution tree the Ethernet link is. It does not matter if the
Ethernet link is point-to-point or multi-access. (Of course, if the
link is not Ethernet, for example a PPP link (RFC 6361), then there
might not be any outer.MacDA.)

> And when the frame reaches to RB4, whether RB send to all nodes in this
> distribution tree with the outer.MacDA is All-RBridges or not?
>
>                                           (RB2)     === (RB4) (tree root)
>                                      //             \\
>                   (Ingress) (RB1)           (RB3)

The TRILL distribution trees are bi-directional trees that reach all
RBridges in the campus. Which RBridge is the "tree root" matters only
when the tree is being calculated. After an RBridge has calculated the
forwarding entries for the tree, it no longer matters what the root
was and the type of forwarding is the same for any forwarded
multi-detination frame. If the frame is output through an Ethernet
port, its outer.MacDA is set to All-RBridges.

Frames do not just go to the root and then get distributed by the
root. In your diagram above, any multi-destination frame ingressed by
RB1 will be sent to RB2. RB2 will send it to both RB3 and RB4. And RB3
and RB4 will not forward the frame as a TRILL data frame. (Any or all
of the RBs may decapsulate the frame and send it locally as native
frame where appropriate.) It make no difference to the actual
distribution that RB4 was the root used by all the RBridges to
calculate the tree.

Thanks,
Donald
=============================
 Donald E. Eastlake 3rd   [phone] (cell)
 155 Beaver Street, Milford, MA 01757 USA
 [e-mail]

> Thanks ,
>
> --
> +--
> | Nguyen Huu Tho.
>    Mail: [e-mail]  |