Synchronization with IEEE 1588 PTP. Terasync. Part 2

Another major problem in this field is the large values of the delay variation. As the existing network node equipment can apply some sort of queuing, the packets that arrive at the node are not distributed forth immediately, but are delayed at least by the time of the packet reception, usually more, as other packets also arrive to the node. The maximum delay can be varied in according to the packet length and the current traffic load of the particular node, as well as by the equipment technical characteristics. The initial concept of the network is a cloud, where there is no pre-defined route from the sender to receiver, so the traffic can propagate through different number of nodes, thus adding to the total delay variation. The existing methods of quality of service (QoS) management don’t provide much improvement.

For example, in 100 Mbit Ethernet the minimum packet length is 64 bytes, which corresponds to the minimum queue delay of 5.12 microseconds. The maximum delay is infinite, as the packet can be dropped (e.g., due to the traffic congestion), but delay variation values in order of magnitude of a hundred of microseconds per node can be realistic. In real networks measurements, the PDV can reach a few milliseconds, and tends to increase to order of magnitude of ten milliseconds when wireless link is used. For comparison, the jitter in legacy E1/T1 networks is in order of magnitude of hundreds of nanoseconds – about 10000 times less. The existing engineering methods for legacy network synchronization can no longer work for the enormous delay variation values of the packet networks. New methods are needed, but their development is seriously impaired by the lack of an appropriate theoretical models.

As no viable theory suitable for the engineering implementation exists, the industry turned to the simulation/emulation approach, based on the investigation of traffic behavior in the existing networks via laboratory experiments or computer emulation of some specific network. The main disadvantage of the algorithms that are developed based on this approach is that the changes in network topology and traffic load patterns can seriously affect the results, sometimes even rendering the previous experiments as unusable. As the modern networks are constantly evolving, it is obvious that this approach can not provide a general answer to the problem. Nonetheless, a lot of the synchronization solutions existing on the market were developed in that way. Therefore, when using such a solution, it is not guaranteed that it will perform sufficiently in a different environment.