U-Turn%20Alternates%20for%20IP/LDP%20Local%20Protection%20draft-atlas-ip-local-protect-uturn-00.txt

1
U-Turn Alternates for IP/LDP Local
Protectiondraft-atlas-ip-local-protect-uturn-00.t
xt

• Alia Atlas (aatlas_at_avici.com)
• Gagan Choudhury (gchoudhury_at_att.com)
• Christian Martin (cmartin_at_verizon.com)
• Brent Imhoff (brent_at_lightcore.net)
• Don Fedyk (dwfedyk_at_nortelnetworks.com)
• Raveendra Torvi (rtorvi_at_avici.com)

2
Outline of Talk

• Overview of Solution
• Control-Plane Modifications
• Data-Plane Modifications
• What Needs to Be Standardized
• Repair Coverage
• Complexity Analysis
• Comparison with Other Methods

3
Loop-Free Alternates Limited Coverage

• The coverage in a network provided by loop-free
  alternates is limited.
• U-turn alternates expand the coverage on real
  networks.
• Analysis on networks shows improvement on average
  from 79.5 to 98.4 coverage of
  source-destination pairs.
• Sufficient to become a network engineering
  problem and not a technology problem (with its
  associated technical complexity).

4
U-Turn Alternates Cooperatively Breaking the
Loop

• R2 can locally determine to use R1 as a U-turn
  alternate if and only if
• R2 is the primary neighbor of R1 for any shortest
  paths from R1 to R4 that go through R2 (R1 is a
  U-turn neighbor of R2).
• R1 has signaled that it is capable of breaking
  U-turns on that interface (traffic received from
  R2 destined to R4 will go to R1s alternate and
  not back to R2).
• R1 has a loop-free node-protecting alternate (R5)
  to reach the destination (R4).

5
Control-Plane Node-Local Computation

• A router S must compute
• For each destination D (via an enhanced SPF)
• If a neighbor N has indicated that it can break
  U-turns for traffic coming in an interface,
• Does that neighbor N have a loop-free
  node-protecting alternate to reach the
  destination D?
• Does that alternate path also avoid the router
  Ss primary neighbor P?
• If a loop-free node-protecting alternate is
  available, select it for use.
• If not, pick among loop-free link-protecting
  alternates and u-turn alternates as desired
  (router-local decision).

6
Control-Plane Routing Protocol Changes

• A router must know
• If a neighbor can redirect U-turning traffic on a
  particular interface
• Interface-wide capability - not tied to
  particular traffic prefixes
• And the policy configuration that neighbor has
  for using its interfaces as alternates.
• Assumes operator has administrative control to
  disallow using an interface as an alternate.
• Signal this information via a new Link
  Capabilities sub-TLV in IGP
• 1 bit U-turn capable recipient
• 1 bit Eligible Alternate
• No additional signaling required based on
  topology changes (i.e. at time of failure or
  after).

7
Data-Plane Mechanisms Rerouting
Destination Incoming Interface Out Interface
R4 L1 L2
R4 L2 L3
R4 L3 L2

• Forwarding Outcome is based on incoming
  interface,
• which is similar to VRFs, RPF checks, ACLs,
• policy-based forwarding, etc.

8
Data-Plane Modifications Encapsulation

• Traffic redirected to U-turn alternates does not
  require any type of encapsulation.

9
What Needs to Be Standardized

• U-Turn Alternates Require
• A Link Capabilities sub-TLV with 2 bits used.
  These Signaling Protocol extensions would be for
  ISIS and OSPF.
• A common selection method for deciding
• To use a loop-free node-protecting alternate, if
  any is available
• How to break ties among those

10
Repair Coverage

• U-Turn Alternates improve coverage on real
  networks.
• Improvement is topology-dependent.
• Minor changes to network can lead to further
  improved coverage.
• Analysis based on source/destination pairs, not
  of traffic covered or of link or node failures
  fully covered.

Id be happy to analyze any network with the
automated tool.
11
Complexity Analysis Control Plane

• Computational Complexity is O(neighbors)
• Neighbors which arent available for use as
  alternates dont count. (They decrease the
  complexity to O(alternate-capable neighbors))
• Feel free to discuss the details after meeting

12
Complexity Analysis Data Plane

• For interface-specific FIBs, no changes required
  - only different information in FIB.
• For non-interface-specific FIBS, need to look at
  the results of forwarding decision and decide
  based on the primary out-going interface and
  incoming interface whether to send traffic to
  primary or alternate.
• Requires additional comparison for determination
• Has potential requirement to read a second
  forwarding result.
• No more look-up complexity than uRPF, VRFs,
  Policy-based forwarding, etc.

13
Comparison with Other Methods

• Commonalities
• Assumes a common framework of alternate
  pre-computation and traffic redirection on
  failure.
• Assumes a base of loop-free alternates.
• Provides a mechanism to break the loop needed to
  go through an upstream node that can provide an
  alternate path.
• Looking for operationally simpler method than TE
  Fast-Reroute
• Computation of alternates may be similar.

14
Comparison with Other Methods Differences

• U-turn Alternates
• Computational complexity is less ( O(neighbors) )
• No encapsulation is needed.
• No set-up, monitoring or maintenance of explicit
  tunnels is required. Particularly important
  because tunnels may need to change when the
  topology does. This can leave protection gaps
  while the new tunnels are created.
• No added complexity or support to learn mechanism
  for directed forwarding.
• No new adjacencies (such as for LDP) need to be
  considered. I.e. a targeted LDP session isnt
  necessary to learn the labels understood by a
  neighbors neighbor. Simply works for LDP.
• Lowest impact on security
• Goal is to simplify operations and provide local
  protection not to make so complex that RSVP-TE
  Fast-Reroute is preferable

15
Conclusion

• U-Turn Alternates offer improved coverage with
• Similar computational complexity to loop-free
  alternates
• No new encapsulation or explicit tunnels
• Simple notification of capability is the only
  signaling extension
• Simple to manage and deploy
• Orthogonal to MPLS
• Comments?