Measuring impact of packet reordering on internet protocol networks

Packet Reordering in IP networks is a phenomenon which is becoming increasingly important in network performance analysis. Reordering is a consequence of network equipment manufacturers increasing switch and link level parallelism within networks, in the quest for performance, reliability and fiscal gains. Wireless technologies are also expected to increase the amount of packet reordering observable in an end-to-end path. This thesis addresses the issue of measuring the impact of packet reordering on Internet traffic, by proposing a number of measurement methodologies and metrics. Previous techniques assume that packet reordering does not often occur, or make assumptions which severely limit the results obtained. This thesis proposes a twopoint passive measurement technique, which improves on previous methods by allowing lightweight measurement of the amount and extent of reordering observed in a TCP flow, and classification of the cause of each reordering-induced packet retransmission. A large testbed measurement study performed using this technique indicated that TCP is tolerant to large percentages of reordered packets, providing that the delay of these packets is maintained below a threshold relative to Round-Trip-Time. This study further indicated that the effects of TCP packet reordering are not always negative. In specific scenarios reverse-path reordering can increase the overall throughput of a flow. This thesis further proposes a mid-point passive Measurement Technique and Visualisation Metric of TCP packet reordering, designed to classify out of sequence packets for many thousands of concurrent TCP flows. This technique is lightweight to implement and does not require symmetric TCP connections to operate. Finally, this thesis argues that future packet reordering metrics must correlate reordering observed at the network layer, with the resulting impacts observed at the application layer. An example of an application-specific metric is developed for MPEG-4 video over UDP traffic, and this metric is used to describe the effects of packet reordering on streamed video traffic.Packet Reordering in IP networks is a phenomenon which is becoming increasingly important in network performance analysis. Reordering is a consequence of network equipment manufacturers increasing switch and link level parallelism within networks, in the quest for performance, reliability and fiscal gains. Wireless technologies are also expected to increase the amount of packet reordering observable in an end-to-end path. This thesis addresses the issue of measuring the impact of packet reordering on Internet traffic, by proposing a number of measurement methodologies and metrics. Previous techniques assume that packet reordering does not often occur, or make assumptions which severely limit the results obtained. This thesis proposes a twopoint passive measurement technique, which improves on previous methods by allowing lightweight measurement of the amount and extent of reordering observed in a TCP flow, and classification of the cause of each reordering-induced packet retransmission. A large testbed measurement study performed using this technique indicated that TCP is tolerant to large percentages of reordered packets, providing that the delay of these packets is maintained below a threshold relative to Round-Trip-Time. This study further indicated that the effects of TCP packet reordering are not always negative. In specific scenarios reverse-path reordering can increase the overall throughput of a flow. This thesis further proposes a mid-point passive Measurement Technique and Visualisation Metric of TCP packet reordering, designed to classify out of sequence packets for many thousands of concurrent TCP flows. This technique is lightweight to implement and does not require symmetric TCP connections to operate. Finally, this thesis argues that future packet reordering metrics must correlate reordering observed at the network layer, with the resulting impacts observed at the application layer. An example of an application-specific metric is developed for MPEG-4 video over UDP traffic, and this metric is used to describe the effects of packet reordering on streamed video traffic