Analysis of UDP Traffic Usage on Internet Backbone Links* Min Zhang Maurizio Dusi Wolfgang John *This study was performed while authors visited CAIDA at UCSD under supervision of kc claffy
Outline Motivation Dataset description Analysis of UDP traffic usage UDP/TCP ratio Per-port analysis Conclusion
Motivation Is there an increase in UDP traffic? Internet fairness and stability concerns Where does additional UDP traffic stem from? P2P applications (e.g. uTP protocol)? IPTV applications? Other unexpected reasons? Study variability in data sets differences of geographical locations, times, networks?
Datasets TimeLengthNetwork location Geogr. location CAIDA-OC hourTier1 backboneUS CAIDA-OC hourTier1 backboneUS GigaSUNET minTier2 backboneSweden OptoSUNET minTier2-Tier1 connection Sweden
Analysis (1) Number of UDP flows has increased since 2002; TCP still carries most of packets and bytes.
Analysis (2) Before 2003: around 40% of UDP flows run on ports below 1024; After 2003:usage of ephemeral ports (>1024) has become common; Today:around 95% of the UDP flows run on ports >1024. top-used ports (in terms of flows): DNS, NTP and NetBios P2P applications 4672, 4665 (eDonkey) 6881 (BitTorrent) 6346 (Gnutella) 6257 (WinMX)
Analysis (3) Top-ten ports: fewer than 7 packets and less than 10KB on average. Larger UDP flows appear mainly in the older traces, suggesting a drift in usage of UDP toward small (signaling) flows.
Summary We investigate the trend of UDP traffic since 2002 a cross networks: from Tier1 and Tier2, different geographical locations Number of UDP flows has increased TCP is still the dominant transport protocol in terms of bytes and packets Most UDP flows use ports > 1024 instead of traditional UDP service ports Flows on ephemeral and P2P ports carry few packets and little data Preliminary conclusion: Current increases in UDP traffic are mainly due to signaling traffic of P2P applications. Next steps? Verifying the preliminary conclusion by more detailed analysis. Continuing to monitor available data, to track trends of UDP usage. Comparing to data from China (UDP-based IPTV is already common). Investigating if UDP patterns can be used as a signature for traces, in order to infer usage patterns?