1 IGP Data Plane Convergence Benchmarking draft-ietf-bmwg-igp-dataplane-conv-app-00.txt draft-ietf-bmwg-igp-dataplane-conv-term-00.txt draft -ietf-bmwg-igp-dataplane-conv-meth-00.txt Authors: Scott Poretsky, Avici Systems Brent Imhoff, WilTel Communications 57th IETF Meeting – Vienna
2 IGP Convergence has become the most heavily discussed Router Benchmark for the Internet Convergence events cause packet loss visible to the network user for most real-time applications Industry Goal <1sec (milliseconds) IGP Convergence – As stated at numerous NANOGs by Service Providers and router vendors. – Applies to IS-IS and OSPF. – Externally observable Black-Box event. Motivation for Draft
3 Convergence Methodologies Control Plane Convergence Data Plane Convergence DUT Update DUT Data Preferred Next-Hop 2 nd Best Next-Hop The point in time in which a single node updates its routing table and reroutes traffic to the new next-hop Data Plane Convergence is in the order of seconds Data Plane Convergence is the major problem to solve Addressed in draft-ietf-bmwg-igp-dataplane-conv-term, meth, and app The point in time in which a single node advertises the route table change to its peer in a route advertisement message Control Plane Convergence is in order of hundreds of milliseconds Addressed in draft-ietf-bmwg-ospfconv- terminology and methodology
4 Basic Test Methodology Test Configuration Tester_1 advertises simulated topology via routing protocol(s) Tester_2 sends traffic to DUT to all destinations in FIB DUT has two paths (via Link 1 and Link 2) to reach destinations DUT by default prefers lower cost path via Link 1 Network Event reroutes traffic to Link 2 Test Procedure Send maximum Throughput with fixed packet size to all destinations in the FIB Measure pps received at Tester Force a Network Event Observe loss of traffic forwarding Observe recovery to forwarding at line rate Calculate Convergence Time Test Setup NOTE: Different test cases Described in draft. DUT Tester Link 1 Link 2
5 Example Convergence Measurement Rate-Derived Convergence Time Route Convergence Recovery Transition Route Convergence Event Transition Packet Sampling Interval Convergence Start TimeConvergence End Time Packet Loss Convergence Time
6 Data Plane Convergence Calculation Calculation 1: –Rate-Derived Convergence Time = Time to converge 100% of the traffic to all routes in FIB after a Convergence Event Calculation 2: –Loss-Derived Convergence Time = (Number of Packets Lost)/(Offered Packet Load per Second) Applying Calculations –Loss-Derived Convergence Time > Rate-Derived Convergence Time –Rate-Derived Convergence Time is Preferred –Loss-Derived Convergence Time must be used when Packet Sampling Interval >100msec.
7 Next Steps – Terminology Draft 01 Add global statement that metrics are externally observable Add new terms: Packet Sampling Interval, Convergence Start Time, Convergence End Time (as shown in Slide 5) Change terms to match those suggested on mailing list (and used in slides 5 and 6): –Full Convergence Time -> Rate-Derived Convergence Time –Average Convergence Time -> Loss-Derived Convergence Time –Route Convergence Event Slope -> Route Convergence Event Transition –Route Convergence Recovery Slope -> Route Convergence RecoveryTransition –Reroute Convergence Time -> Route Convergence Time (Definitions are not changing – only terms) … the ready for Last Call for IESG Review?
8 Next Steps – Methodology Draft 01 Add Reporting Format with Packet Sampling Interval, IGP Timer values, fixed packet size, and FIB size used and results for Rate-Derived Convergence Time and Loss-Derived Convergence Time. Add suggested values for IGP Timers to “Test Considerations” section Add “Guidelines for Packet Sampling Interval” to “Test Considerations” section as discussed in slide 6 … then ready for Last Call for IESG Review?