1. 1 Variability in TCP Round-trip Times Jay Aikat Jasleen Kaur Don Smith Kevin Jeffay Department of Computer Science University of North Carolina at Chapel Hill The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL http://www.cs.unc.edu/~jasleen/research/RTT

2. 2 TCP Round-trip Times (RTTs) Popular belief: RTTs do not vary significantly within TCP connections Mean RTT can be used to approximate per-segment RTTs Mean RTT can be used to approximate per-segment RTTs TCP throughput models TCP throughput models [Altman00, Kumar98, Lakshman97, Mathis97, Padhye98, …] Analysis based on TCP behavior Analysis based on TCP behavior RTT of initial segments is typical for a TCP connection RTT of initial segments is typical for a TCP connection[Jiang02] Do per-segment RTTs vary significantly within a TCP connection?

3. 3 Data Source UNC Campus Rest of the world Monitor ISP router Rich data source! DAT ACK UNC Regional health-care center Departments Research labs Departments Research labs Residence halls Wireles s users VPNs Modem VPNs Modem ibiblio web servers

4. 4 Extracting Valid RTT Samples Guiding principle: Guiding principle: Consider only those RTTs where there is unambiguous correspondence between an ACK and the DAT that triggered it. Consider only those RTTs where there is unambiguous correspondence between an ACK and the DAT that triggered it. Caveat: delayed ACKs Caveat: delayed ACKs Could add 200 – 500 ms to RTT estimates Could add 200 – 500 ms to RTT estimates ack [i] dat [i+1] ack [i+2] Monitor Remote end-point dat [i+2] dat [i] ? ?

5. 5 Trace Statistics Connections Remote hosts RTT samples BytesPackets All connections 22.7 million 962 K 252 million 628 GB 511 M Connections with at least 10 samples 1.1 million 258 K 236 million 581 GB 464 M Large data set!

6. 6 Variability Across Connections 60% connections see min RTT less than 100 ms 60% connections see min RTT less than 100 ms Only 23% see max RTT less than 100 ms Only 23% see max RTT less than 100 ms ACKs can arrive more than 25 s after DAT transmission! ACKs can arrive more than 25 s after DAT transmission! Mean and median RTTs are comparable measures Mean and median RTTs are comparable measures

7. 7 Variability Within Connections Median RTT: Median RTT: 30% of connections see a median RTT more than twice the min RTT 30% of connections see a median RTT more than twice the min RTT 90% RTT: 90% RTT: 22% of connections see a 90% RTT more than 5 times the min RTT 22% of connections see a 90% RTT more than 5 times the min RTT 90% RTT increases with min RTT 90% RTT increases with min RTT

8. 8 The SYN / (SYN+ACK) RTT RTT yielded by the SYN and SYN+ACK pair RTT yielded by the SYN and SYN+ACK pair Differs by more than 10% from min RTT for 14% of connections Differs by more than 10% from min RTT for 14% of connections Differs by more than 10% from median RTT for 50% of connections Differs by more than 10% from median RTT for 50% of connections

9. 9 Per-Segment RTTs: Mean or Distributions? Is mean RTT a good approximation for per-segment RTTs? Is mean RTT a good approximation for per-segment RTTs? TCP analytical models TCP analytical models TCP evaluation (simulations) TCP evaluation (simulations)

10. 10 Ongoing Work Impact of RTT variability on past work Impact of RTT variability on past work TCP analytical models TCP analytical models Delay-based congestion control Delay-based congestion control TCP evaluation (simulations) TCP evaluation (simulations) TCP-based analysis TCP-based analysis Causes of variability Causes of variability Congestion? Congestion? End-hosts? End-hosts? Models for per-connection RTTs Models for per-connection RTTs Accurate simulation environments Accurate simulation environments