Measurement and Classification of Out-of-Sequence Packets in a Tier-1 IP Backbone Sharad Jaiswal, Gianluca Iannaccone, Member, IEEE, Christophe Diot, Jim Kurose, Fellow, IEEE, Don Towsley, Fellow, IEEE Paper Presented by Saroj Patil 2/23/2019 CS622

Introduction Study of out-of-sequence packets in TCP connection over Sprint IP backbone. Measurements taken in the middle of end-to-end path. Analyze traces from OC-12 and OC-48 links in nearly 7600 unique ASes. Result shows relatively consistent rate of out-of-sequence packets of approximately 4% 2/23/2019 CS622

Related Work Benett sent bursts of ICMP probe packets through public exchange point, and reported packet reordering by specific switch. Paxson examined packet reordering and replication in TCP between pairs of measurement sites. Bellardo described technique that can estimate packet reordering in TCP data transfers from a remote host. 2/23/2019 CS622

Methodology Retransmission: Sender infers packet is lost and retransmits the packet. Network duplication: A nonsender retransmitted copy of packet is observed. In-network reordering: The network can invert the order of two packets. 2/23/2019 CS622

Retransmissions R1.a: W’≠W R1.b: t’-t>RTO R1.c: Duplicate acks>3 R2.a: t’-t”>RTO R2.b: duplicate acks>3 and timelag>RTT R3.a: flag infastRecovery set and x<sndHigh 2/23/2019 CS622

Reordering and Duplicates R4: t’-t”<RTT R5: t’-t<RTT 2/23/2019 CS622

Unneeded Retransmissions R6: ACK is lost between measurement point and the sender, or sender timeouts prematurely. R7: Unknown 2/23/2019 CS622

Estimating RTT RTT helps to compute RTO and to identify reordering and duplicates. Sender’s RTT=d1+d2 Sources of Estimation Inaccuracy Variable delays along end-to-end path if d2>d1 overestimate or d2<d1 underestimate observed RTT. End-host delays Estimation of cwnd 2/23/2019 CS622

Experimental Validation Three sets of experiment with no packet drops, 3% drop rate and 5% drop rate. Also another set with wireless hop. Relative Error of RTT Estimation pi is latest packet observed at measurement point. Spi and mpi are the most recent samples taken at sender. n is the total number of packets observed. 2/23/2019 CS622

2/23/2019 CS622

Measurements and Classification Research Measurement Data Out-of-sequence Packets Network Anomalies 2/23/2019 CS622

Discussion of Results Representativeness of Sampled Flows Compare number of ASes. Examine the distribution of number of path length. Compare frequency. 2/23/2019 CS622

Measuring End-End Reordering/Duplication Frequency A duplicate ACK may be generated upon receipt of packet classified as out-of-sequence. A receiver sends duplicate ACKs upon receipt of the unneeded transmission Duplicate ACKs are also used to update the receiver advertised window If the receiver follows a delayed ACK policy, some of out-of-sequence-events will not trigger a duplicate ACK. 2/23/2019 CS622

Conclusion About 4% of packets generated by TCP connections are out-of- sequence, most of which are due to retransmission in response to a packet loss. Packet reordering affects about 1-1.5% of all data packets, however, they have little impact on the quality of a TCP/IP protocol connection as perceived by the end users Other network anomalies such as duplication of packets represent a very marginal phenomenon in the Internet. 2/23/2019 CS622

Future Directions Identify causes behind packet losses i.e. congestion, routing or link failure Study if a TCP connection experiences congestion in a single or multiple bottlenecks along the path Identify tools to detect ASes that are responsible for the out-of-sequence packets. 2/23/2019 CS622

References [1] J. Bellardo and S. Savage, “Measuring packet reordering,” presented at the ACM SIGCOMM Internet Measurements Workshop San Francisco, CA, Nov. 2002. [2]P. Benko and A. Veres, “A passive method for estimating end-to-end TCP packet loss,” presented at the IEEE Globecom Taipei, Taiwan, Nov. 2002. [3] J. Bennett, C. Partridge, and N. Shectman, “Packet reordering is not pathologial network behavior,” IEEE/ACM Trans. Networking, vol. 7, Dec. 1999. [4] R. Cáceres, N. Duffield, D. Towsley, and J. Horowitz, “Multicast-based inference of network-internal loss characteristics,” IEEE Trans. Inf. Theory, vol. 45, pp. 2462–2480, Nov. 1999. [5] C. Fraleigh, S. Moon, B. Lyles, C. Cotton, M. Khan, D. Moll, R. Rockell, T. Seely, and C. Diot, “Packet-level traffic measurements from the sprint IP backbone,” IEEE Network, 2003. [6] S. Jaiswal, G. Iannaccone, C. Diot, J. Kurose, and D. Towsley, “Measurement and classification of out-of-sequence packets in a Tier-1 IP backbone,” presented at the IEEE INFOCOM, San Francisco, CA, 2003. [7] S. Jaiswal, G. Iannaccone, C. Diot, J. Kurose, and D. Towsley, “Inferring TCP connection characteristics through passive measurements,” presented at the IEEE INFOCOM, Hong Kong, 2004. [8] H. Jiang and C. Dovrolis, “Passive estimation of TCP round-trip times,” ACM Comput. Commun. Rev., vol. 32, no. 3, Jul. 2002. [9] V. N. Padmanabhan, L. Qiu, and H. J. Wang, “Server based inference of internet link lossiness,” presented at the IEEE INFOCOM, San Francisco, CA, 2003. [10] V. Paxson, “End-to-end routing behavior in the internet,” IEEE/ACM Trans. Networking, vol. 5, pp. 601–615, Oct. 1997. [11] ——, “End-to-end internet packet dynamics,” IEEE/ACM Trans. Networking, vol. 7, pp. 277–292, Jun. 1999. [12] L. Rizzo, “Dummynet: A simple approach to the evaluation of network protocols,” ACM Comput. Commun. Rev., vol. 27, no. 1, Jan. 1997. 2/23/2019 CS622