IEEE1588 and Switches

The precision of the protocol also depends on the latency jitter of the underlying network topology. Point to point connections provide the highest precision, with hubs imposing very little network jitter. Under very low or no network load, Layer 2 switches have a very low processing time, typically 2 to 10µs plus packet reception time. In this case, new switch designs also have low latency jitter e.g. the Hirschmann Switch RS2-FX/FX with about 0.4 µs latency jitter.

But switches are working with queues and store and forward, so only one queued maximum length packet imposes a delay for the following packet of about 122µs, and under high load conditions, more than one packet will be in the queue. The next issue for the precision of the protocol is that latency is completely symmetric for both directions: from the master to the slave and visa verse. This can nearly never be guaranteed under higher network loads.

Prioritization of packets e.g. IEEE802.1D/p does not really solve the problem, because at least one long packet can be in front of a synchronization packet and so will impose up to 122µs to the jitter of transmission. Currently, available switches show that after the priority scheduler, there is another queue for 2 up to 8 packets, and not only one as expected. This means a jitter from 360µs up to 1ms under heavy load conditions.

The solution for all these problems is the usage of IEEE 1588 Boundary clocks in switches. In this case you only have point to point connections and there is nearly no delay jitter between master clock and slave clock and internal queuing delay/jitter of switches is not relevant any more.