Methodology for Testing High-Speed Network Devices with Predicted Traffic Patricia Sagmeister IBM Research Zurich www.zurich.ibm.com Motivation Validation & performance analysis of network devices Alternatives Synthetic traffic Actual traffic Test traffic requirements Reflect actual traffic characteristics High-speed networking traffic Worst-case conditions Combination of actual traffic characteristics with generation methods www.zurich.ibm.com Methodology OC-3 Link Trace Fully Loaded OC-3 Link Trace Trace 1 Time -compressed Link Trace Trace 2 Trace 3 Final Trace www.zurich.ibm.com Testsuite Selection Testsuite Traces Real Traces Traffic Modifier Mod. Traces Traffic Analyzer Network Device Testsuite Selection Traffic Analysis Results www.zurich.ibm.com Testsuite Characteristics Absolute No. of Flows 350 Relative Protocol Distribution acc. to Pkts 300 120 250 No. of Flows Thousands % of Protocols 100 TCPPkt 80 UDPPkt 60 200 FlowNo 150 100 ICMPPkt 40 50 20 0 Traces 0 Traces Simultaneously Active Flows Relative No. of Pkts <= x Bytes 30 120 25 Nr of Flows Thousands % Pkts <= x Bytes 100 80 60 40 20 20 MaxActiveFlows 15 MinActiveFlows 10 5 0 0 Traces PktLength<1520B PktLength<600B PktLength<200B PktLength<48B Traces www.zurich.ibm.com Testsuite Characteristics Link 0 5 Combined Links 3 5 2 4 1 0 Time Steps in ms NoNewPkt Thousands Thousands 4 3 2 NoNewFlows 1 Link 1 0 3000 Time Steps in ms 2500 NoNewPkt 2000 NoNewFlows 1500 1000 500 0 Time Steps in ms NoNewPkt NoNewFlows www.zurich.ibm.com Conclusion Based on real traffic, monitored today, we have presented a method to extrapolate future high-speed traffic behavior Step 1 : compacting of real traces -> leads to a full loaded link Step 2 : time compression -> leads to a high-speed trace The resulting trace has the same traffic characteristics as the original actual trace, but shows additional phenomena that represent serious challenges for a network device Higher period of rhythmic fluctuations in packet size within a flow Higher period of fluctuations in the occurrence of hosts and their local share Interleaving of the time-compressed traces leads to longer traces. Aggregated high-speed network traffic is obtained through random or one-by-one interleaving. Concatenation of the same trace increases the number of simultaneously active flows Additional interleaving method: Step 1 : time correct interleaving of different real traces Step 2 : compacting & time compression www.zurich.ibm.com