6.6 Rules of Thumb for Packet Timing Network Implementation

At the time of writing, the delay variation effects in complex networks have not been studied thoroughly enough for estimating delays at the fine-grain accuracies required for timing purposes. A network can be assessed by carrying out delay measurements over multiple connections at different load conditions. Unfortunately, just after building the network the loading is still much less than it will be after some time. However, for example MAFE and pktfilteredMTIE produce quantitative results where the margin between the limit and delay variation can be observed. If at a later measurement the margin has become clearly smaller, then some links might need upgrading.

Experience so far indicates that if the packet timing slaves are good enough, there is no need to decentralize time servers to the edges of the transport network. The path in Figure 6.31 is part of a commercial deployment. There are two copper/fiber Ethernet links followed by ten packet microwave radio (MWR) links.

Figure 6.31 Example of a path in a commercial MWR backhaul deployment.

Although the implementation above has worked well, it cannot be taken for granted that all such implementations will work. An example set of rules for 16-ppb clock accuracy target and G.823 2-Mbit/s traffic mask is given below. The values match with a packet clock that has a 10000-s MAFE corner point shown in Figure 6.25.

• Maximum one way delay should be < 100 ms.
• Jitter < 5 ms.
• Packet loss < 2%.
• Clock packet stream should have the highest priority or at least the same priority as the real-time traffic and receive expedited forwarding QoS.
• High-priority traffic share of BW should be 60 % or less.
• Maximum number of hops: 20.
• Maximum number microwave hops: 10. In this case the total number of hops should be less than 15.
• The average load of the links along the path should not be persistently above 50% if the path is long.
• The number of delay jumps should be limited to a few per day

For targeting clock stability requirements including long observation intervals, such as G.824 1.5-Mbit/s mask or G.823 SEC mask, clearly stricter rules apply. The ITU test case setup of ten hop chain with 1-Gbit/s links should be acceptable.

The long MWR chain described above utilizes FDD (Frequency Division Duplex) technology where both directions of the links are active continuously enabling good packet timing performance. However, there are also some TDD (Time Division Duplex) radios on the market. In this case each direction is waiting for half of the time causing potentially problems in packet timing. Especially, more than one TDD link in a row could be risky for packet timing.

SHDSL (Single-pair High-speed Digital Subscriber Line) and VDSL (Very-high-bit-rate Digital Subscriber Line) have survived as mobile backhaul links at the same time as PDH is becoming obsolete. Neither DSL technology has notable issues in traversing packet timing except for the lower bit rate compared with point-to-point optical interfaces.

Passive Optical Networks (PONs) operate continuously in downstream but employ TDM (Time Division Multiplexing) technique upstream. Therefore, the delay floor in upstream direction is quite noisy and downstream direction is usually smoother.