Keeping order: determining the effect of TCP packet reordering

Packet reordering over TCP/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 on the Internet, seeking performance, reliability and economical improvements. This paper presents a methodology for simulating and measuring TCP reordering, providing an insight into the behaviours of the congestion and retransmission algorithms, and demonstrating that reordering has a measurable effect on performance. These measurements illustrate that there is a maximum reordering delay threshold that should be applied to packets, regardless of percentage reordering, below which reordering has negligible effects. Determination of this threshold, on a specific path, is key to ensuring that a specific switch or router does not introduce reordering to such an extent that it causes unnecessary retransmissions and an associated reduction in throughput.