Agenda

1
Lecture 2

• Agenda
• Finish with OSPF, Areas, DR/BDR
• Convergence, Cost
• Fast Convergence
• Tools to troubleshoot
• Tools to measure convergence
• Intro to implementation scheduling
• Readings
• Sub-milisecond IGP convergence
• uLoop elimination

2
To READ

• Sub-millisecond convergence
• Threads vs. events

3
HWks

• Study the scheduling or Quagga
• Fast convergence in quagga OSPF
• LSA generator
• Implement BFD
• We will take it slow

4
Some complications

• EXAMPLE
• In a broadcast network can not have n2
  adjacencies
• Use a fake centralized router
• Designated router, one is elected as such
• Setup adjacencies only with it
• The DR advertises the network LS for the network
• Backup-DR to ensure that if DR dies I recover
  quickly

5
Scalability concerns

• EXAMPLE flooding cost and number of routes
• AREAS
• Limit the scope of flooding
• Limit the number of routes
• If allocated hierarchically/properly
• Area border router (ABR) between two areas
• ????PLE how to compute areas through ABR
• Area 0 is special and is the backbone
• Stub areas that do not have through traffic

6
Periodic LSA refresh

• To catch some rate memory corruptions
• Necessary to make the protocol really robust
• In OSPF refresh each 30 minutes
• Synchronization of updates
• If not careful all routers will refresh all their
  LSAs at the same time
• Randomization

7
What matters

• Convergence speed
• How quickly all routers will have consistent
  information after a change in the network
• uLoops cause me to loose traffic
• How quickly new routes start to being used
• So traffic flows properly again
• Stability
• How much protocol control traffic
• How much CPU I burn
• How things work when CPU is overloaded
• The above may be conflicting goals

8
Important Times in IGP

• EXAMPLE timeline
• Link fails, router detects failure, sends update,
  computes SPF, updates RIB, updates FIB
• Failure detection time
• Depends on link technology
• May need to rely on the HELLO protocol
• Flooding time
• Depends on the CPU load and network load,
  interfaces
• SPF time
• Depends on how loaded is the CPU and how many
  routes I have
• Depends on the algorithm
• RIB/FIB update
• How fast I change forwarding plane
• depends on number of routes

9
Network wide timeline

• Routers next to the failed node will detect the
  failure, originate LSA
• LSA will travel the diameter of the network
• Last router will compute SPF
• Last router will install routes in FIB
• Done

10
How to be faster

• Faster SPF
• Better algorithms
• Incremental SPF
• Faster detection
• Faster HELLOs
• BFD!!!
• In the line card instead of the control plane
• many protocols can share
• Faster FIB download
• Download important prefixes first
• Do things faster
• Trigger SPF immediately
• Trigger LSA origination immediately

11
How to be stable

• SPF may be expensive
• Can not do SPF all the time something minor
  changes, may be better to do one SPF for all
  changes
• Avoid extra FIB downloads
• Dot not want to do SPF all the time if there is
  network churn, will overload CPU
• Do not want to sent too many updates at once
• Receiver may get overloaded
• Do not want to send updates too quickly
• Link may be flapping
• When CPU/links are loaded ensure that
• Do not miss HELLOs, will make things worse

12
Configuration Timers

• Hello timer, dead timer
• LSA update delay
• LSA pacing
• LSA retransmission pacing
• SPF delay
• Wait for this time before you do SPF
• SPF hold-time
• Do not do another SPF before this time passes
• Can have dynamic timers
• Be fast when CPU is idle
• Be slow when CPU is loaded

13
How to measure performance

• Black box vs. white box
• White box is near impossible for commercial
  products
• Black box needs tricks
• Without knowledge of the internal-structure
• See paper