IP Flow Performance Measurement (IPFPM) Framework draft-chen-ippm-coloring-based-ipfpm-framework-03 Dallas, Mar 2015, IETF 92 Mach Chen Ed. Vero Zheng Ed. Greg Mirsky Ed. Giuseppe Fioccola Ed.
Motivation Why IPFPM? ü Measurement rely on injected packets has limitations ü Operators needs to know the performance of the real traffic running in their networks What Can IPFPM Do? ü IP Flow Performance Measurement ü Measure the real traffic ü Support MP2MP flow measurement, with data correlation and reorder tolerant mechanism ü Enable SLA verification, fault localization and fault delimitation
IPFPM Components Data Report (IPFIX) MCP Collects the measurement data from the Measurement Agents (MAs) and calculates the performance metrics according to the collected measurement data MA1 MA2 Measurement Agent (MA) MA3 MA4 Measurement Agent (MA) Executes the measurement actions (e.g., marks the packets, counts the packets, records the timestamps, etc.), and reports the data to the Measurement Control Point (MCP) Period Number ü Data collected correlation
Packet Loss Measurement Packet Loss ü Use one or more unused bits to mark the packets ü Different markings ( 0, 1 ) divide the flows into different consecutive blocks ü Counting based on each block of marker, two counters, one for 0 block, the other for 1 block Tx 5 Zeros 5 Zeros No packet lost Rx Flow A Flow B A A A A A A A A A A A A A A R1 Marking Changing B B B B B B 5 Zeros B B B B B B B 4 Zeros B R2 One packet get lost
Packet Delay Measurement Only marking ONE packet in a period, Measure the packet delay based on THE packet Time synchronization is required Marking and Record Timestamp Record Timestamp Tx Rx Flow R1 R2
Average Delay Measurement l Marking EVERY packet of each block l Calculation of average timestamp for each block (sum of all the timestamps divided by the total number of packets) l The Average Delay between two nodes is calculated by subtracting the related average timestamps l Reduce bias of sparse sampling l One bit for both loss and delay Tx Calcula<on of Average Timestamp for all packets of each block Calcula<on of Average Timestamp for all packets of each block Rx Flow A A A A A A A A A A A A A A A R1 Marking Changing R2 Note: draft-tempia-ippm-p3m-00 as reference
MP2MP Flow Measurement Flow ID:1 Flow ID:2 1 1 1 1 1 U- MA1 MCP Measurement results Flow ID:1 U- + 2, D- Count = 4, no packet loss Flow ID:2 U- + 2, D- Count = 3, one packet lost D- MA3 1 1 1 1 1 A 2 2 2 2 2 Des<na<on 1 Source 1 Flow ID: 1 Count: 4 1 1 1 1 1 U- MA2 D- MA4 Source 2 2 2 2 2 2 2 2 2 2 2 Des<na<on 2 Flow ID:1 Flow ID:2 Flow ID: 2 Count: 3 Period number = Read <me modulo Time period The counts and timestamps from distributed MAs are indexed by period number and flow ID
MP2MP Flow Measurement Flow ID: 1 MCP Flow ID: 1 Count: 4 1 1 1 1 1 U- MA1 D- MA3 1 1 1 1 1 A 2 2 2 2 2 Flow ID: 1 Flow ID: 2 1 1 1 1 1 U- MA2 D- MA4 Flow ID: 2 Count: 3 Flow ID: 2 2 2 2 2 2 2 DCP-B 2 2 2 2 Period number = Read <me modulo Time period A flow can be monitored at multiple upstream and/or downstream MAs ü Flow 1 and Flow 2 are monitored at U-MA1 and U-MA2 ü Flow 1 s destination is D-MA3, flow 2 s destination is D-MA4
Re-ordering Tolerance Each MA maintains two timers (C-Timer and R-Timer with the same interval) ü C-Timer for changing color ü R-Timer for reading count and timestamp, in order to allow for a certain degree of packets re-ordering ü R-Timer should be started later than delta-t after C-Timer started ü t1 < (R-Timer + delta-t) < t2 T4 T3 U- MA T2 T1 D- MA Block 3 Block 2 Block 1 Block 3 Block 2 Block 1 A A A A A A A A A A A A A A A A A A A A A A A A T4 T3 T2 T1 t2 t1 Safe sec<on to read count of Block 1
Progress Was draft-chen-coloring-based-ipfpm-framework Became draft-chen-ippm-coloring-based-ipfpmframework-00 at IETF 87 Berlin, after IPPM re-charter ü -00 new author from Vodafone and CNNIC ü -01 new author from China Telecom -02 Split into two documents at IETF 90 Toronto ü IPFPM Framework (method, reference model, etc) ü IPFPM Report (IPFIX information elements, templates and optional templates, etc) ü Structure change, terminology Change ü New author from Ericsson
Update Since Last Version New author: ü Giuseppe Fioccola (Editor), Telecom Italia Reference Model: ü Refer to LMAP Reference Model Delay measurement: ü Adding Average Measurement Editorial Changes
Summary Stable document, 03 revision Straightforward mechanism, flexible measurement A whole system, fault tolerant, data correlation Favored by multiple operators Running code, real implementation
Next Steps WG adoption Reviews and comments always welcome