Delay characterization through in-protocol measurements

We present an in-protocol Internet delay measurement technique for web browsing. By using the handshake and tear-down phases of a TCP protocol exchange as timers, we can make estimates of the round trip times without using an external tool, command or ICMP. During a web page request and retrieval, the latency encountered is typically caused by a combination of the end-terminal and network processing delays. One proportion of the delay, deterministic in nature, arises from the protocol processing or the speed of light propagation. Another proportion of the delay, more stochastic in nature, arises from interaction with cross traffic, queuing in buffers or web server response times. Our approach in this work is to clearly separate the sources into deterministic or stochastic types of delays. In order to evaluate our in-protocol measurement approach, we have identified two scenarios in two environments. The first scenario is the delays as seen by a client only on the public Internet, this is the most representative scenario from a user perspective. To quantify delays in more detail we have measured the delays at both a client and a server in our own research environment. We call this environment CheesePi, which is a Raspberry Pi measurement infrastructure. It currently runs over the SUNET Swedish academic network. It is important in this second scenario we can access both clients and servers in several locations.