Hopefully, by this point in this series, the reader will have a good understanding of IEEE 1588 technology and the ways it can be applied. It should be clear from the discussions of applications in test and measurement, industrial automation, and telecommunications that this is a very active field of investigation, involving many companies and researchers around the world.

However, it is far too early to be sure that IEEE 1588 will achieve widespread adoption, or that the additional capabilities it brings to time-based operations will prove sufficiently useful to displace or augment existing technologies.

It is clear that IEEE 1588 has the potential to make a major contribution to applications with hard real-time constraints. It is probably not too much of a stretch to claim that IEEE 1588 will be successful in the field of industrial automation and motion control.

The area of test and measurement is less certain, and depends as much on how quickly the distributed architecture and peer-to-peer communication model exemplified by the LXI specification is accepted, as it does on the IEEE 1588 standard itself.

Telecommunications applications at this point must be considered speculative. There are clearly areas within telecommunications where IEEE 1588 has the potential to make significant contributions, but it is not at all clear whether it will ultimately be adopted.

Proposed Techniques to Enable IEEE 1588 in Telecommunications
The telecommunication network impairments that degrade the performance of an IEEE 1588 system are latency fluctuations and asymmetry. If only frequency alignment is required, then asymmetry and the absolute value of the latency are of no concern. However, if time transfer—epoch alignment—is also required, then all impairments must be considered.

Asymmetry. Asymmetry impairments can arise from a variety of causes. Optical fiber systems supporting two-way communication will have different path lengths in the two directions as a result of chromatic dispersion, even if the nominal line lengths are identical. In practice, companies already correct for this effect by means of additional fiber on the short path. This is essentially a calibration process.

Far larger contributions to asymmetry can be expected from queuing effects in switches. In addition, some network operators use ring topologies in which traffic flows in only one direction, which leads to vastly different path lengths for the forward and reverse communications between two arbitrary points. There are also protocols that are asymmetric by design, e.g., asymmetric digital subscriber loops (ADSL).

If these factors can be rendered time-invariant by network engineering techniques, then the needed corrections can be made by calibrating the resulting network. This is not very appealing, but may be the best that can be done.

It remains to be determined how well asymmetry can be controlled in actual operating environments. Algie estimates that the latency across a typical metropolitan area network is less than 20 ms. If this is the case, then asymmetry-induced epoch offsets would be on the order of at most 10 ms, which is probably not good enough for many of the proposed telecommunications applications.

As discussed later, early data are discussed indicating that in some circumstances the latency is substantially less. If it were possible to transfer time to the edges of the metropolitan networks using existing equipment, then it may be possible to use current IEEE 1588 techniques, such as boundary clocks, to distribute time within an enterprise.

The most difficult environment for controlling asymmetry for the telecommunications examples discussed earlier in this series is the metropolitan area network, and the backhauls to wireless cell sites. In building distribution or distribution within equipment racks, the networks are much more likely to have controlled and stable asymmetry properties.

Latency Fluctuations
Latency fluctuation impairments arise from a variety of sources. The most common is queuing fluctuations in switches. These are particularly troublesome because they are likely to be very traffic-dependent. A second source of timing fluctuation is the equipment needed to translate between communication protocols present in the network, e.g., the transition between a TDMA and a FDMA protocol, or from a T1 line to SONET.

On a longer time scale, changes in path routing will also introduce differences in latency. Furthermore, these fluctuations may not be the same on the forward and reverse paths, which will further aggravate the asymmetry problem.

The only practical solution to fluctuation is filtering the offset and delay values computed by the slave clocks. Simple filtering is not likely to be effective due to the time and traffic load dependence of the magnitude and distribution of the delay values.

More sophisticated algorithms could monitor the distributions, and make use of the holdover properties of the local oscillator to maintain a stable time scale while the parameters for a new distribution are computed.

It will probably be necessary to separately monitor the forward and backward paths, since there is no reason to assume that the effects of traffic will be uniform in the two directions.

The observation times for these filtering algorithms will be long. If we assume that the fluctuations are uniformly distributed and have a magnitude of about 2 ms, then the NTP curve suggests that to achieve an Allan deviation of 10^-8, observation times of about 1 day will be required.

The early data discussed later indicate that observation times on the order of a few hours may suffice. Observations times on this order can be supported by quartz oscillators. If this is not the case, then more stable oscillators such as rubidium will be required, which will increase the cost and make the use of IEEE 1588 in these applications much less attractive.