SDN: new flow control capabilities in mesh networks

Hello, dear readers. It should immediately make a reservation that this article is not about what is good and what is bad in SDN or some other network technology. We are not violent adherents of software-configured networks. We just want to tell you about the solutions we came to when developing industrial mesh networks as part of creating industrial wireless communication systems . Tell about the opportunities that are at the junction of technology, allowing you to rely on well-proven solutions and at the same time keep up with the times.

The efficiency of data transmission networks is largely determined by the network-level algorithms implemented in them (routing, load limiting, transmission of service information). The number of such algorithms is quite large, which is caused by the variety of requirements for information exchange, the features of the transmitted information flows, as well as the capabilities of hardware and software. In addition, each of the protocols has its own advantages and disadvantages, and, apparently, there is simply no way for an ideal routing algorithm, and it cannot exist.

This is fully applicable to mesh networks where two routing approaches prevail - proactive (for example, the OLSR protocol) and reactive (the AODV protocol). There are also very productive hybrid schemes.
')
OLSR (Optimized Link-State Routing) [1] is based on the collection and distribution of service information about the state of the network. As a result of processing this information, each node can construct a model of the current state of the network in the form of a formal description of the graph, the vertices of which are assigned to the network nodes, edges (or arcs) to communication lines (links). Having such a graph, any node can calculate the “lengths” of the shortest paths to all destinations in the network and choose the “optimal” route leading to any particular network node.

This algorithm responds well to many unforeseen events, which, first of all, include:

• spontaneous failures / restoration of nodes and lines;
• damage and repair of network nodes;
• aggressive effects of the “external environment” leading to blocking of individual elements of the system;
• connection and disconnection of nodes and lines during operational relocation of subscribers.

Application of the OLSR protocol is most effective in networks with a high density of nodes, in which traffic of irregular, sporadic nature prevails. OLSR constantly uses some bandwidth for its service traffic.

Along with the OLSR protocol, which provides a preliminary (proactive) calculation of routes between all pairs of subscribers within a subnet, another network is implemented in mesh networks — the reactive AODV (Ad hoc On-Demand Distance Vector) protocol [2]. This protocol provides for the search and organization of routes as they are needed, and the destruction of found routes after their use.

The AODV protocol uses less resources than OLSR, because the size of service messages is relatively small and requires less bandwidth to maintain routes and routing tables. But at the same time the number of connections and disconnections of nodes in the network should be small.

Nevertheless, both protocols (in aggregate), to some extent, meet the basic requirements that the “ideal algorithm” must satisfy [3]. These requirements can be formulated and commented as follows:

Correctness. Both algorithms are efficient and do not contain logical contradictions.

• Computational simplicity. The implementation of the AODV algorithm requires minimal processor resources. And, conversely, the OLSR algorithm is rather complicated.
• Sustainability. Both algorithms tend to a specific solution without sharp oscillations with insignificant changes in the states of radio waves.
• Justice. At first glance, both algorithms provide equivalent services to all users of the network. But, due to its topological location in the network, individual users can capture a relatively large share of resources than they need.
• Adaptability to changes in traffic and topology. Both algorithms are able to determine the new set of routes when the network topology changes. But he and the other algorithm is static with respect to external traffic and load distribution.

The latter drawback is characteristic of all known decentralized strategies , with the exception of the shortest-queue and offset strategies [3], in which the route weight is determined as a linear function of two parameters: the number of transit sections on the way to the addressee and queue lengths on the directions of delivery lying on the route ( The classic example is the Internet [3]). There are no such strategies in mesh-network protocols, and, apparently, it will not, since such a strategy will require complex and time-consuming computations at all nodes and will hardly have the property of stability - minor fluctuations in the load can provoke route recalculation.

Centralized strategies are free from these shortcomings, which imply a global distribution of service information from an external "intelligent" device (controller) to all nodes and vice versa - from all nodes to the controller, and can be adaptive with respect to changes that are reported in the service content. posts.

In its simplest form, the controller calculates route tables using any of the optimal flow distribution methods using the topology known to it and the specified external traffic. At the same time, among the other possible two approaches can be noted: the controller periodically distributes the tables calculated by it across all nodes and the virtual channel approach — calculating individual routes for each pair of subscribers. The second approach can be very useful and even necessary when organizing data transmission in real time.

The advantage of centralized strategies is the elimination of intelligent computing on all nodes except the controller, the main drawback is the reliability of the network is determined by the reliability of the controller and the secure channels used to control the slave nodes. When the controller fails, the network either ceases to function at all, or will function inefficiently - dynamic routing will turn into static. In addition, any centralized strategy implies the possibility of delivery of proprietary information to all network nodes, which is not always possible when it is deployed and reconfigured in an aggressive or degrading network environment. It is also worth noting the fact that the controller's reaction time to events of managed nodes is very important. If the controller's response time includes an assessment of the situation, decision making, generation of control actions and their delivery to the nodes. If it is less than the speed of occurrence of control events, the network inevitably degrades.

A compromise approach is also possible - a combination of centralized and decentralized strategy. Under favorable conditions, a centralized strategy functions, ensuring optimal distribution of the most important flows using a controller, and when a controller fails, routeing and other intelligent calculations are performed using available decentralized algorithms.

This approach can also be implemented in mesh networks, but it requires the use of specialized routing and addressing tables that specify the direction of transmission of packets not only to the addresses of the recipients, but also to more specialized features, for example, dedicated pairs - “source-recipient”. In standard network protocols, there are no such possibilities, but this problem is rather trivially solved using the OpenFlow protocol.

One of the unobtrusive, but the most significant features of this protocol is that it allows you to understand almost anything as a “stream” by creating your own classifier of streams adapted to a specific network task. For example, in the first version of the OpenFlow protocol, the TCP stream may be determined by data from 10 fields. OpenFlow allows you to find a compromise between standard solutions and unique algorithms that can coexist in the same environment without “interfering” with each other.
The essence of this compromise is as follows:

• mechanisms are built into the standard protocol stack that allow dividing the serviced flow into parts, one of which can be served by standard TCP / IP protocols, and the rest - using original procedures;
• For sorting, the service fields of the received packets can be used, such as the IP addresses of the sender and the receiver, the corresponding MAC addresses, port numbers, etc.
• To determine the routes of “non-standard” flows, flow tables can be used that determine the correspondence of packet transmission directions not only to recipients' addresses, but also on the basis of more complex dynamic logic — source-recipient, access rights, server loading, service priorities, etc. .

In the conclusion of this article, it is worth noting that the hybrid approach most clearly manifests itself in the construction of networks operating in conditions of limited resources, harsh operating conditions. Such networks include industrial communication networks, special-purpose networks, rapidly deployable communication systems, as well as wireless mesh networks.

If this article arouses interest in the community, we are happy to write a sequel in which we will try to highlight the issues of implementing the described hybrid approach in industrial wireless mesh networks.

Thanks for attention!