Latency requirements determines the degree of user satisfaction about the web service. Software development practices shall consider the end-to-end latency as one of user-oriented characteristics during the entire life-cycle of an application. Latency budget defines latency target and non-functional requirements for distributed systems.
We have learned that resolution of human-friendly names into network addresses is a pre-requirement of any Internet service transaction. Therefore, latency of the DNS infrastructure contributes to the overall user-experience. A slow DNS gives the impression that the service is slow.
TCP/IP do not utilizes a full bandwidth when connection is established. It gradually increases a data flow. Effect of TCP/IP slow start and its latency becomes visible for applications while mobile network is used.
The radio access networks are slow and applications needs to optimise number of round trips.
Network is a system that make peers to wait. One-way delay shows the network performance from source to destination. Measure the delay and optimise protocols for number of round-trips.
Propagation delay is 5 microsecond per kilometer due to speed of light constrain. The packet handling time of hardware-assisted network elements is in range of 4-20 microsecond, on averages 25 microsecond per hop. There is not any solution to impact on latency except to minimize the distance required by the signal to travel, measure and deal with this latency.
The network infrastructure involves various technologies, interfaces and protocols along the communication path. It is conveniently partitioned into domains: Internet, Data Center and Internet Service Provider. The post gives an introduction to high-level IP network architecture and its impact on communication latencies.