The all-IP trend and the explosive growth of mobile data services have imposed higher demands on the wireless backhaul networks that require capital-intensive investments in order to keep up with the ever increasing backhaul capacity.
Network Operators are eagerly researching ways to minimize this investment. A business procedure to do that, is via network sharing. Even in this scenario, where two or more operators share the backhaul network, an increase in the bandwidth utilization and thus the efficiency in the mobile backhaul network must be researched and implemented.
Network Management Challenges
Improving the efficiency of the backhaul network requires an active involvement of network management. But this is not the only challenge for the network management of mobile backhaul networks.
Network management has become more important also due to the nature of packet-based transmission and the resiliency and flexibility features of packet networks.
Therefore, the management of wireless backhaul networks encounters the following challenges:
- Management of microwave networks employing different technologies – such as PDH / Hybrid / Native-Packet Point-to-Point, Point-to-Multi Point, etc. – and handling multiple service types of different requirements.
- Provisioning of services end-to-end while ensuring Quality of Service (QoS), providing the same OAM experience as with traditional SDH and MSTP networks.
- Monitoring of resources and services end-to-end, especially for services transported over multiple interfaces (E1-TDM, E1-IMA, Fast Ethernet, etc.) and through Pseudowire (PW) technology at the last-mile and the aggregation segments of the backhaul network.
- Adaptive Modulation Coding (AMC) which allows links to use higher modulation under good weather condition, to increase spectrum efficiency. Thus, the network bandwidth is becoming dynamic.
- Introduction of Small Cells, that has made the network topology more complex and difficult to manage.
Till now, the answer to management challenges has been the Unification of Management, i.e. to use a single platform to manage multiple technologies from multiple vendors.
But is this enough, especially to handle the increased network complexity imposed by Small Cells and Adaptive Code Modulation?
The Adaptive Modulation challenge
This in effect makes microwave radio links to transport services with different availability. In the above example at least 40 Mbps of bandwidth is usable (i.e. available) regardless of mode, but additional 40 Mbps can only be used when link uses 16QPSK or higher. Thus, the differences in the availability of modulations can be applied also to bandwidth and therefore service flows. In our example, 40Mbps of bandwidth has higher availability than 155Mbps.
Till now, backhaul networks were planned and rolled-out with simple tree-or star topologies. This topology provides one single path for service flows, thus planning must make sure that the links are used efficiently. The problem that arises due to AMC, is that the network bandwidth becomes dynamic; therefore it is no longer effective to try to maximize efficiency during planning.
NMS is the Solution
As bandwidth utilization cannot be efficiently maximized during planning, due to Adaptive Modulation Coding, we need a solution that will also be dynamic in nature.
One potential solution is to introduce a new advanced function from the NMS, for automatic traffic re-routing. With this function, NMS will be triggered by backhaul links modulation changes, in order to calculate and re-route traffic to a different path, to maximize efficiency in bandwidth utilization.
In order to do so, the network topology must be adapted as well to allow multiple traffic paths. The current network designs of mobile backhaul networks consist of tree, star and some ring topologies.
To enable multiple paths, tree topologies must be transformed to mesh. But are all mesh networks the same?
Mesh networks and link density
To find the theoretical maximum number of links, we use simple mathematics as following: for a network consisting of N nodes, the theoretical maximum number of links that can be created in a mesh network is N*(N-1)/2. This maximum is reached when all nodes are interconnected, thus each of the N nodes is connected to the rest N-1 nodes.
For example, in a small network consisting of 7 nodes, if we implement a full mesh architecture we can have a theoretical maximum number of 21 links (=7*6/2). So, if we implement only 8 links, then the network link density will be 8/21=0,38.
Let’s see another interesting example to understand why link density is important. In a medium size network consisting of 50 nodes, we see that the maximum number of links is 1225 (=50*49/2). This is a very big number that in real life will never be implemented due to investment cost. A more realistic scenario would be to implement 74 links, arranged in a tree topology, with maximum link length = 5 km, scattered around a 30 × 30 km region. In this case the link density 74/1225=0,06.
It seems obvious that the higher the link density, the more capable the NMS will become to re-route traffic to maximize bandwidth utilization. But as we see in the two above examples, the bigger the network, the smaller the link density that is can be achieved in the real world.
So how do we select which nodes to interconnect?
Transforming tree topologies to mesh
Not really. In real life, the probability that two nodes are interconnected decreases with the link distance. This is because big link distances requires are more difficult to implement and for example for wireless backhaul this would mean bigger antennas (thus cost) and possibly decreased availability.
The formula used to express with mathematics the probability of two nodes to be interconnected, is following
p = α exp(−d/(β dfix)),
- α = 0.2, β = 0, 0.05, 0.1, 0.2, 0.4
- d = distance between two nodes, dfix = 15 km
How to re-route traffic in mesh networks
Now that we have de-mystified the concept of mesh networks, let’s try to understand how to perform traffic re-routing, with an example.
In the below example network, we have implemented six links to backhaul traffic from two eNBs to the metro network (the ring at the right hand side). Take a look at link #6. This is the link that transforms this tree network to mesh, as it interconnects the two eNBs, increasing the link density of the network.
Let’s try to calculate the link density for this small network:
- We have 6 nodes
- We have a maximum of 15 links (=6*5/2)
- We have implemented 6 links.
So, the link density is 6/15=0,4.
Let’s assume now that we want to backhaul 3 services with a total of 125Mbps traffic.
- The active path to transport these 3 services is the one marked gray in the diagram that consists of the links numbered: “3-2-1”. We call this route A.
- As we see in the diagram, there is an alternative route, consisting of the links “6-5-4-1”. We call this route B.
Now let’s assume that due to weather conditions, Link #3 has changed modulation from 128QAM to QPSK. The available bandwidth of the active route A has decreased from 155Mbps to 40Mbps, and as a result, some services will become unavailable or will have degraded performance. At the same time, we may have route B underutilized, i.e. bandwidth utilization to be 40% of 155 Mbps.
As NMS monitors all modulation changes in real time via threshold alerts, it will detect the modulation change at link #4. What we want now to happen, is NMS to automatically re-route the services above 40Mbps bandwidth to pass through route B (6-5-4-1).
How can we do that?
The traffic re-routing Algorithm
The pre-requisite to perform traffic re-routing, is to be able to provision end-to-end services.
Typically, end-to-end service provisioning uses the Dijkstra algorithm that is efficient at defining the shortest path in terms of hops given a source/destination pair. For Dijkstra, a path with 2 hops is preferred from a path with 4 or more hops, i.e. Djikstra algorithm does not take into account bandwidth utilization or availability of links. Thus in our example, the Dijkstra algorithm will only select route A, as it has less hops compared to route B.
Let’s now try to add constraints for availability.
Setting a minimum availability constraint to Shortest Path
We start with a minimum availability threshold, i.e. a link with anything below 99.995% to be deemed as unusable. In this assumption, we consider the availability of usable bandwidth as the availability of the link.
Note that we know that
- QPSK can carry 40Mbps,
- 16QAM can carry 80Mbps,
- 32QAM can carry 108Mbps,
- 128QAM can carry 155Mbps.
If we now run Djikstra, it will calculate for each route the number of hops. We can easily exclude routes that include a link with availability less than 99,995%.
Can we further enhance this algorithm?
Adding end-to-end Availability constraint to Shortest Path
The next enhancement is to calculate the end-to-end availability for each route, and select the route that:
- Has the higher availability.
- Can carry the required bandwidth.
We can also select a secondary stand-by path, using this method. So, if we return to our example, once the NMS is notified for a modulation change, it will calculate for the two routes the end-to-end availability and will re-route service accordingly.
Reallocating bandwidth limitation
The algorithm described is a definite improvement over Shortest Path, but has an inherent problem. It selects and “populates” routes in a first-come-first served procedure, i.e. services provisioned later may not be able to find a route available to satisfy their availability and bandwidth requirements.
In this case we need to:
Reallocate bandwidth of a flow to lower availability route if the flow can maintain required end-to-end availability.
Improving the algorithm with Partial Disjoint Protection
Π Sn (1 – (1 – Π Wn) (1 – Π Pn))
– Sn: Set of shared links
– Wn: Set of links used only in working path
– Pn: Set of links used only for protection path
Partial Disjoint Protection is used in optical networks with success, therefore their use in mobile backhaul is also recommended.
- To increase network efficiency when admitting flows, it is recommended to use an algorithm that calculates and uses end-to-end availability, i.e. it is better to find path that meets end-to-end availability requirement rather than shortest-path routing.
- Additionally, having a protection path can improve network efficiency, as low availability bandwidth which are not usable for single path can be used because of the protection path.
- Bandwidth reallocation is effective when difference of availability of bandwidth within a same link is large while there is little flexibility when difference is small
- Effectiveness of each method varies with topology and link conditions
- Partial Disjoint Protection is more effective compared to 1+1 protection when topology is not fully meshed
Benefits of traffic re-routing
- We increase the effective capacity of wireless links to maximize usage of the available infrastructure by dynamically setting the paths based on the network conditions and status.
- We minimize hassle for network operator, as the re-routing will become almost instant and automatic.
- We use the most efficient path under current network conditions for the services
- We correct traffic imbalances, in a smart and automatic way.