Achieving Convergence-Free Routing using Failure-Carrying Packets K. Lakshminarayanan et al. Presented by Ang Li 06/29/07

Convergence is the source of evil! Packet Loss LSA storms (high control message overhead) –High CPU utilization –Network instability What we have –Loop-free convergence –Fast convergence –Backup paths

FCP, no convergence at all! No control message exchange Routes calculated “on-demand” –When a packet meets a failure Packets carry topology information –Failure list –To avoid loop

FCP algorithm

Why FCP works? A stable and consistent Network Map –FCP falls back to normal link-state protocol when no failure –Guarantee efficiency Failure list grows monotonically –Guarantee loop-free

FCP with Source Routing To deal with inconsistent Network Map –FCP packets carry the whole path besides failure list –Differences in Network Map are reported as “failures” –Higher packet overhead

Two Properties of FCP Guaranteed Reachability –Packets are delivered within finite time if the network is still connected –Time is bounded by number of failures and network diameter Path Isolation –Routers not on the forwarding path of one packet can not affect the forwarding

Reducing Overhead of FCP Two overheads –Computation time –Packet overhead

Reducing Overhead of FCP Two overheads –Computation time Solution: precompute “secondary nexthop” & caching –Packet overhead Solution: mapping between failure list and label

Network Map Dissemination Centralized scheme –Need special care to guarantee consistency Otherwise FCP-SR is used

Pros of FCP Theoretically interesting –A new approach –Provide two layers of topology update Network Map layer Failure layer Security –Influence of a hijacked router is limited

Cons of FCP Far from practical deployment –Many details missing For e.g., how to do caching? How much caching is enough? –Inconsistent with today’s router architecture Route computation can not be carried out in linecards Idea is not new –Already in wireless context –The only thing new is the concept of Network Map