Packets and latency

We all know what IP packets are. They deliver data across a network, be it a file transfer, VDI session or VOIP packet. It’s a package.  An application is usually a bi-directional flow of packets, but the individual packets only flow in one direction. There may be a corresponding packet flowing in the return direction, for example a TCP ACK or the other half of a VOIP conversation, but packets are uni-directional commodities. So why do we think of latency in terms of a round trip? Surely the only thing that matters is the latency in the direction the packet is going? Historically we think of IP networks as symmetrical in nature. I.e it’s the same latency in each direction. This may be true in LAN environments. It may also be true in some WAN environments such as MPLS, but it is by no means guaranteed. Latency is rarely symmetrical in the Internet, so any mechanism that uses round trip latency to compare the latency of WAN paths is fundamentally flawed, because you cannot derive the one way trip time from that measurement, and that is the important factor for an IP packet, because after all, packets are uni-drectional in nature.

Categories: Software Defined WAN (SD-WAN)