TRILL issue: How many/which trees?

1
TRILL issue How many/which trees?

• Radia Perlman
• Radia.Perlman_at_sun.com

2
Issue

• Nice to have multiple trees
• More optimal delivery
• Avoiding out-of-order deliver
• Multipath multicast
• Ingress RBridge chooses which tree

3
Why would it matter to have per-source spanning
tree?
S
11
6
X
7
9
3
2
5
10
4
14
4
RBridges need to know all the trees

• To precompute ports in the tree
• To add ingress RBridges to RPF list

5
How many trees?

• Easiest way to decide RBridge with highest
  (priority, ID) tells everyone
• LSP contains
• My tree priority
• How many trees I want computed
• This same tree priority is used to determine
  which trees will be computed

6
So, now we know how many trees but which ones?

• The highest priority RBridge could enumerate all
  of them, but that would take up a lot of space.
  Instead it just says n
• All potential trees are sorted by (priority, ID).
  The top n are the choosable trees.

7
In the LSP

• Currently
• Total trees just the number n
• Trees I might select a list of nicknames, or if
  empty, means I will only select the top priority
  tree
• Proposal to save room
• Have a number k for Trees I might select, which
  means Ill take the top k out of n
• And optionally, a list (if some reason to want a
  different set of trees)

8
Why would anyone care about which trees?

• An RBridge might want trees with roots closer to
  it
• Or it might want trees that are as disjoint as
  possible, for multipathing

9
Really minor change being suggested

• Currently total trees only specifiable by n
• Do we want it optionally to be enumerated?
• Or partially enumerated, as in n, with these k
  specific ones Ill enumerate, and the rest of the
  n being the top (priority, ID) roots
• And it should be omittable
• Individually choosable trees now only
  specifiable by enumeration (or exactly one, the
  top one)
• Do we want that optionally to be just k?
• Would it be a different TLV than total trees?
• I think yesthough it might be nice to be able to
  omit n