Understanding when to deploy PMP
One of the most common questions we’re asked by new customers is,
“How do I decide when to deploy point-to-multipoint (PMP) and when to deploy point-to-point (PTP) for my backhaul?”
This is a great question, because the two technologies are very complementary to each other.
An engineering perspective
From an engineering perspective, in making the choice between PMP and PTP for a given link, we are seeking to maximise efficiency and utilisation of the equipment and the RF channel while satisfying a set of requirements for throughput, latency and link availability.
In economic terms, this translates into choosing the technology which gives the lowest total cost of ownership while satisfying those requirements.
An excellent way to make the choice between PMP and PTP is to look at the characteristics of the data traffic we want to carry.
I’m going to consider mobile broadband backhaul traffic here, because that’s what the majority of our customers use our technology for today.
In a future post I will talk about small cell backhaul traffic.
As ever, the NGMN has some useful information we can use, in the whitepaper Guidelines for LTE Backhaul Traffic Estimation.
This paper describes (§2.2) the initially counter-intuitive result that the peak throughput for an eNodeB actually occurs, not during busy hour, but during quiet time. This is because:
“During busy times, there are many UEs being served by each cell. The UEs have a range of spectrum efficiencies, depending on the quality of their radio links. Since there are many UEs, it is unlikely that they will all be good or all be bad, so the cell average spectral efficiency (and hence cell throughput) will be somewhere in the middle.
During quiet times however, there may only be one UE served by the cell. The cell spectrum efficiency (and throughput) will depend entirely on that of the served UE, and there may be significant variations … the scenario under which the highest UE and cell throughputs occur [is]: One UE with a good link has the entire cell’s spectrum to itself. This is the condition which represents the ‘headline’ figures for peak data rate”.
Figure 4, reproduced here, illustrates this point:
The paper goes on to give the following peak and mean traffic figures for a number of LTE configurations:
Understanding the peak-to-mean ratio
What we can immediately see from this figure is that the peak-to-mean ratios of the traffic in the dominant, downlink, direction are very large, ranging from about 4:1 to almost 6:1.
This agrees well with measurements we see from real networks.
For example, the following from a very busy HSPA+ network show the peak-to-mean ratio of the backhaul traffic for each node B (on the y-axis) plotted against the peak backhaul demand for that node B on the x-axis.
When traffic has a high peak-to-mean ratio like this, we call it “bursty”, as opposed to “smooth” when the peak-to-mean ratio is close to 1.
Data traffic in general, for example on LANs and residential internet access connections as well as mobile networks, is bursty; and this presents a difficulty in carrying it efficiently on PTP links, as shown here:
The problem here is that a PTP link with a single traffic source (a ‘tail link’ in the backhaul network) needs to be dimensioned to carry the peak traffic, but there is only a single source of offered load.
Therefore the utilisation of the link (or efficiency) is equal to the mean offered load divided by the capacity, or in other words the reciprocal of the peak-to-mean ratio of the traffic.
So if my traffic has a peak-to-mean ratio of 4:1, the maximum utilisation of a PTP link carrying that traffic is ¼, or 25%.
In the chart above, you can visualise this as all the white space below the red line being wasted bandwidth, which is provisioned but unused.
It’s important to say that this is not a failing in PTP systems in any way – it is simply that the characteristics of the traffic are not well suited to the static bandwidth provisioning that PTP provides.
The advantage of a PMP system is that it can serve multiple sources of offered load simultaneously.
The bandwidth of the shared RF channel is dynamically allocated to different sources as required.
Conceptually, then, the peaks and troughs from different traffic sources ‘cancel out’ to some extent, as we illustrate in the following live network example showing eight nodeBs being backhauled by a single VectaStar Gigabit sector.
Here we are also relying on another property of the traffic, namely that peak demands for different nodeBs do not occur at exactly the same time.
We discuss this at greater length in The Effect of System Architecture on Net Spectral Efﬁciency for Fixed Services.
Liberating spectrum to meet growing capacity demands
A useful analogy here is to think about a bank with deposit accounts.
Banks operate a fractional reserve system, meaning that they are only able to repay a defined fraction of the total of deposits at any given time.
This therefore relies on the observation that, statistically, not everybody goes to the bank and withdraws all their savings at the same time.
When this assumption breaks down, there is a ‘run on the bank’.
In a similar way, we rely on the observation that, statistically, not every node B requires its theoretical peak backhaul throughput at the same time.
When this assumption breaks down, things are a bit less dramatic however – we simply discard some low priority traffic.
This is perceived (if at all) by users as a temporary reduction in internet browsing speed.
Crucially, we can dimension the system in such a way as to set the probability that this occurs to a value of our choosing.
The advantage of fractional reserve banking is that it liberates dormant capital for further investment and lending.
Likewise, the more efficient use of RF channels in PMP systems liberates dormant electromagnetic spectrum (provisioned but unused, as in the example above) for use addressing the ever-growing capacity demands of modern mobile networks.
In conclusion, then, some brief rules of thumb for when to deploy PTP and when to deploy PMP are as follows:
|… when traffic is smooth (voice dominated)
|… when traffic has already been aggregated
|… in the middle mile of backhaul
|… for long distance links
|… when spectrum is uncongested or inexpensive
|… when traffic is bursty (data dominated)
|… to create an on-air traffic aggregation
|… for tail links (last mile)
|… for dense deployments
|… when spectrum is congested or expensive