WO2014009104A1

WO2014009104A1 - A method for defining a domain within a network

Info

Publication number: WO2014009104A1
Application number: PCT/EP2013/062494
Authority: WO
Inventors: Martin Johnsson
Original assignee: Waterford Institute Of Technology
Priority date: 2012-07-11
Filing date: 2013-06-17
Publication date: 2014-01-16

Abstract

A method for defining a domain within a network of nodes mutually interconnected across respective network links is disclosed. The method comprises for each of a plurality of nodes of the network, determining within a network neighbourhood comprising a plurality of tiers T_n of nodes surrounding the node, each tier T_n comprising nodes within a given number of links n of the node, a sum of nodes Sum_nodes(T_n) in each tier. For each of the plurality of nodes, a connectivity score CS is provided, the score being a function of the sum of the products of the sum of nodes Sum_nodes(Tn) and the tier number n for each tier. Based on the connectivity scores for the plurality of nodes, the extent of at least one network domain within the network, is determined the or each domain comprising a plurality of domain nodes associated with a centroid node having an optimal connectivity score within the plurality of domain nodes.

Description

A method for defining a domain within a network

Field

The present invention relates to a method for defining a domain within a network. Background Management of networks such as, 4G LTE is a challenge and large-scale networks typically require a clustered network structure for efficient data collection and dissemination.

Yuanzhu Peter Chen, Arthur L. Liestman, and Jiangchuan Liu, "Clustering Algorithms for Ad Hoc Wireless Networks", in Ad Hoc and Sensor Networks, ed. Y. Pan and Y. Xiao, Nova Science Publishers 2004 surveys several clustering algorithms for ad hoc networks, concentrating on those that are based on graph domination, with a view to building a virtual backbone for a network to enhance network quality of service.

Stephen. S. Yau and Wei Gao, "Multi-hop Clustering Based on Neighborhood Benchmark in Mobile Ad-hoc Networks", Elsevier Mobile Networks and Applications (MONET), vol. 12, 2007, pp. 381-391 constructs multi-hop clusters with balanced sizes, based on a neighborhood benchmark (NB) which quantifies the connectivity and link stability of mobile nodes and where the nodes with highest NB scores are selected as clusterheads.

Summary

According to the present invention there is provided a method of defining a domain within a network according to claim 1. The present invention can be used to self-organize the formation of domains in an arbitrary network topology such as for wireless mesh networks but equally for fixed networks or even networks having a mix of fixed and wireless networks.

Embodiments of the invention do not require explicit domain configuration of network nodes and node software/firmware implementing the invention may run continuously to cope with changing network topology. The convergence time of implementations of the invention can be extremely fast, being directly proportion to maximum number of node hops and the sum of the link delay, due both to the simple calculations involved, as well as the short messages needed to distribute information between network nodes for the purposes of performing the invention.

Brief Description of the Drawings

Embodiments of the invention will now be described, by way of example, with reference to the accompanying drawings, in which:

Figure 1 shows an exemplar network topology where each node is allocated a unique number used as a name tag;

Figure 2 shows the calculation of CS according to a first embodiment of the present invention and the elected centroid for the network topology of Figure 1; Figure 3 shows the calculation of CS according to a second embodiment of the present invention together with Primary Centroids for the network topology of Figure 1 , where NN = 5, DR=2; and

Figure 4 is a flow diagram showing the operation of an embodiment of the invention Description of the Preferred Embodiment Referring now to Figure 1 , which shows a portion of a network comprising a plurality of nodes 1...65 mutually interconnected by network links - these can be wired or wireless links. Nodes are aware of the network topology at least to the extent of a network domain (partition) within which the node may lie. According to the invention, each node runs an agent (or comprises a distributed component of management overlay) which is arranged to obtain from and provide to at least the nodes of the network within the domain, information comprising a score, to enable an individual agent to determine if its host node is a centroid within a domain of the network. In the present specification, the score is referred to as a Centroid Score (CS) and this reflects the connectivity of a node to other nodes in a network neighbourhood surrounding the node. By calculating and exchanging with each other their CS, nodes are able to identify which node can be elected as the centroid for that domain, the centroid being the node representing the center of a defined domain. The calculation of CS can be done in several different ways as explained below.

Common to each approach is that each node within the network defines their own tiered view of the network topology. The tiered topology (TT) for a node associates the node with

interconnected nodes, each tier indicating a distance corresponding to the number of hops to a node in the network topology. Figure 1 shows the first four tiers TO...T3 of the tiered topology for each of nodes 3 and 45.

In a first embodiment, when a tiered topology has been established for a node, the node can calculate its CS as follows: CS = SUM_TIERS =(SUM_TI + SUM _T2 + . . . SUM_TN,), SUM_Tn=n* SUM_NODES(TN)

Expressed in words, the calculation is so that for each tier, the number of nodes in that tier is calculated (SUM_nod_es(Tn)), and this is multiplied by the tier number (n) to provide SUM_Tn, and finally the sum of all tiers is calculated to provide SUM-ners. This formula generally prioritizes connectivity between the node and the neighboring nodes; with a node with more close neighbors being favored over a node that might even have fewer tiers but which has most of its neighbors far from the node. It will be seem that this particular measure of CS is extremely simple to calculate, disseminate and compare. It will also be appreciated that the CS as represented in the equation above is simply one way of providing this measure and alternative calculations can be employed to provide the score. Figure 2 shows the CS and number of topology tiers around each node within the domain comprising the network of Figure 1.

In this first embodiment, the node with the lowest CS is deemed as being the centroid and in the example of Figure 2, node 34 is chosen as the centroid.

In the embodiment of Figures 1 and 2, by letting all nodes in a domain comprising the entire network topology exchange their respective CS, the nodes in the network topology are able to collectively determine which node has the lowest CS and thus both elect and also let all nodes in the network topology be aware of the node which is the centroid. In case two or more nodes have the same CS, the following criteria could be used to determine a single centroid amongst nodes having equal CS: a) Lowest number of tiers in their topology;

b) Lowest number of nodes in outermost tier of their topology;

c) If the number of nodes in the outermost tier is equal, repeatedly comparing the number of nodes in the next inner tier for each tier until the innermost tier (Tl);

d) Node with "highest" name.

In the first embodiment, nodes become aware of the extent of the tiered topology surrounding them and thus the extent of their network domain when they receive CS messages back from the nodes of the network. For example, in the network of Figures 1 and 2, the tiered topology of the domain extends for 13 tiers around node 28; whereas the topology extends for 8 tiers around nodes 33 and 34.

However, in other implementations of the invention, it may not be desirable, for example, because of the extent of messaging required to implement the invention, for domains to span the entire network topology.

A second embodiment of the present invention extends the first embodiment for the purpose of creating multiple network domains within a greater network topology.

The second embodiment introduces the concept of Network Neighborhood (NN) and Domain Radius (DR) for the purpose of identifying a plurality of network domains within the network topology.

In the second embodiment, each node is pre-configured with a DR and the NN. The DR corresponds to the number of tiers that will comprise a Network Domain (once formed), and the NN is the number of tiers through which the CS for any node will be calculated as well as disseminated. NN could correspond to the known part of the complete network topology for any given node. This provides for a more scalable approach to network domain formation where a trade-off is made in regard of the accuracy of the centroid position, but also to calculate a CS where the local topology takes precedence over the global topology. By default the NN shall be at least twice that of DR + 1. By comparison with the calculation of the first embodiment, in the second embodiment, a normalization factor is introduced, as nodes being further into the centre of a network topology will generally have more connectivity and thus many more nodes within their tiered topology, specifically when taking specific notice of that the scope is restricted to N. Thus the calculation of CS becomes:

CS = SUMxiers SUMx odes,

SUM_nodes(Tl) + SUM_nodes(T2) + · · · SUM_nodes(Tn)

Again other methods of performing the above calculation may be employed. As before, a centroid is then selected based on the calculated CS. This means that if a node finds itself having the minimum CS within the range of nodes corresponding to DR, it will be elected as a centroid. A domain will then be formed by the centroid and comprises the centroid node itself, and at least initially, all nodes being within the range DR of the centroid.

In the example of Figure 3, DR is set to 2 and NN to 5. Thus, for example, the SUM_Tie_rin the 5 tiers surrounding node 10 is 74 and SUM_Tnod_e is 24, giving a CS for node 10 of 3.08.

As a result, nodes 10, 32 and 36 are therefore elected as centroids and form respective domains. These centroids are referred to as Primary Centroids as they can be elected without any further constraints than the criteria mentioned above.

A node which has become part of a domain with a Primary Centroid is not eligible to become a centroid. At the same time, the approach of the second embodiment can initially leave a number of nodes outside established domains, D1-D3 of Figure 3. This gives rise to the possibility of electing Secondary, Tertiary Centroids etc. Such centroids are elected because they do not fall within the domain of a Primary Centroid and while there are nodes with a lower CS within the DR of the node, they are already part of a domain.

Thus, a node such as node 55 can't become a Primary Centroid as the neighbouring node 65 has a lower CS. But as node 65, which has a lower CS is within the DR of node 36 (which is a Primary Centroid), it can't become a centroid. This allows node 55 to become a secondary centroid.

Once such a node determines it has become a centroid, it can emit a message within at least its DR and up to its NN that it has become a centroid. If such a message received by any other non- centroid nodes within the DR of the new centroid, they can decide if they wish to join the domain of the new centroid. So if the new centroid is closer to those nodes than the original (primary centroid), the closer nodes can join the domain of the new centroid. This is a decentralized approach where domain formation is more implicit and it is up to each node to understand and make a selection of which domain it wishes to join.

Other implementations of the invention can be centralized, where an elected Centroid will a) broadcast that it is the Centroid, b) make a selection of which other nodes that will be part of its network domain, and c) as part of selecting nodes to become part of its network domain, negotiate with other Centroids for their nodes - based on the same criteria as for the de- centralized approach. In the Centralized case, there is a need to disseminate Centroid and domain formation information across the NN as peering Centroids need to understand the distance between themselves for proper domain formation.

Whether the centralized or de-centralized approach is taken, in the topology of Figure 3, nodes 19 and 20 are within the DR of the Secondary Centroid in node 29 and the Primary Centroid in node 32, but as these nodes are closer to node 29, they move to the domain formed by the

Centroid in node 29. On the other hand, node 35 is within the DR and at equal distance between the Primary Centroid in node 32 and the Secondary Centroid in node 53. As node 32 has lower CS, node 35 will belong to the domain formed by that centroid.

Similarly, when node 36 starts telling its NN that it is a centroid, node 55 will then discover that node 65 is within the DR of that elected centroid. As a result node 55 can make itself eligible to become a Secondary Centroid (i.e. there are no nodes within the DR of node 55 which has a lower CS and which are NOT within the DR of an already established domain).

Now, for example, node 65 is closer to node 55 than node 36 (both being centroids and node 65 is within the DR of both centroids). Thus, on receiving an indication that node 55 has become a centroid, node 65 will become part of the domain formed by the centroid in node 55.

As secondary centroids become elected, this can in turn lead to tertiary centroids being elected. Thus, node 63 becomes a tertiary centroid (counted as tertiary as it has a relation to a Secondary Centroid in node 52) and node 57 on receiving an indication that node 63 has become a centroid will become part of the domain formed by the centroid in node 63.

In a third embodiment of the invention, the second embodiment is extended for the purpose of merging a network domain at any given level of centroid hierarchy (if the invention is implemented recursively as described below) with another domain.

For example a network domain formed with the second embodiment might be deemed as being too small in terms of number of nodes. Instead of letting such domains operate independently, the third embodiment allows them to merge with another network domain. In an example, a minimal number of nodes for a network domain might be set as three. In the example network of Figure 3 , where DR=2 and NN=5 , there will be three network domains comprising just two nodes, i.e. the domains made up of nodes 57 and 63, nodes 54 and 61, and nodes 27 and 38 respectively.

In this embodiment, which works equally with both centralized and de-centralized centroid selection, any node which is part of a network domain but who wishes to merge with another network domain marks itself as being "unhomed" and requests to join an existing network domain. It then selects a network domain with the centroid for the domain being the closest to the node itself, AND for which there exists connectivity to that network domain. If there are two or more centroids at the same distance and for which connectivity exists (possibly beyond DR but within NN), the centroid with the lowest CS is chosen and if these are equal, criteria a)...d) above can be applied.

This means that the domain comprising nodes 57 and 63 will join the domain formed by the centroid in node 50, and analogously nodes 54 and 61 will join the domain formed by the centroid in node 55, and nodes 27 and 38 will join the domain formed by the centroid in node 36. In this example, the end nodes (63, 61, and 27) can't join until the inner "unhomed" nodes (i.e. 57, 54, and 38 respectively) have joined the domain at which time connectivity comes to existence. As indicated, where domain merging according to the third embodiment is permitted, a domain might have member nodes that are farther away than DR. A node will stay "unhomed" until the node can once again be part of a domain that fulfills criteria to become a domain on its own right. An implementation of the third embodiment is illustrated in Figure 4. At step 50, a node calculates its CS, as in the second embodiment, and broadcasts this within N, step 52. The node in turn accumulates CS calculations from other nodes with NN, step 54. Based on the accumulated CS information, the node can determine if it can become a centroid, step 56. If so, it can broadcast this, step 58, although in some implementations, nodes could independently determine which nodes around the network would become centroids based on the their knowledge of network topology. If a node becomes a centroid, no further action is necessary, until there is a change in network topology, e.g. a new CS score is received, step 64, a new centroid is elected, step 60, or a node is disabled (not shown). If a node does not become a centroid, then it decides if it is to be part of a centroid domain, step 59. If a node cannot be a centroid or part of a domain, it then recursively determines if it can be a subsidiary (secondary, tertiary...) centroid and/or member of a subsidiary centroid domain. Whenever a node becomes a member of a domain, it can determine if the domain contains too few nodes and if it should merge with another domain as described above, step 62. In parallel, the node may received a notification of the election of a new (subsidiary) centroid, step 60, and then depending on how close the node is to this centroid or the original centroid, it may join the domain of the new centroid, step 61.

It will be appreciated that the second and third embodiments of the invention can also operate in a recursive manner where the centroids representing formed domains can form a hierarchy of domains. Thus, centroids are regarded as neighbours if they are interconnected by the shortest number of links and a centroid to centroid link is then equivalent to a node to node hop at the bottom of the hierarchy. Thus domains can be formed of centroids within a given number of hops (DR) from other centroids, to provide "super-centroids" and so on up the hierarchy.

It will be appreciated that the present invention is primarily concerned with defining domains within a network, each of these domains including a centroid which is related to nodes within its domain according to the connectivity score. However, once a domain is defined, the centroid need not necessarily be used as a clusterhead or for aggregating and relaying information to and from nodes of the domain. This responsibility (along with the required resource and/or energy consumption) can be delegated, rotated or shared between any node of the domain according to any appropriate criteria.

It will be appreciated that the present invention is operable within any kind of network including and not restricted to wireless mesh networks, self-organizing networks (SON) or ad hoc networks.

Claims

Claims:

1. A method for defining a domain within a network of nodes mutually interconnected across respective network links, said method comprising the steps of: for each of a plurality of nodes of the network, determining within a network neighbourhood comprising a plurality of tiers T_n of nodes surrounding the node, each tier T_n comprising nodes within a given number of links n of said node, a sum of nodes Sum_nod_es(Tn) in each tier; for each of the plurality of nodes, providing a connectivity score CS, said score being a function of the sum of the products of the sum of nodes Sum_nod_es(Tn) and the tier number n for each tier; and determining based on said connectivity scores for the plurality of nodes, the extent of at least one network domain within said network, the or each domain comprising a plurality of domain nodes associated with a centroid node having an optimal connectivity score within said plurality of domain nodes.

2. A method according to claim 1 wherein said network neighbourhood is limited to NN tiers from each node.

3. A method according to claim 2 wherein said connectivity score for a node is inversely proportional to the sum of nodes SumT_Nodes in said neighbourhood within NN tiers of said node.

4. A method according to claim 3 wherein each domain is at least initially limited to a domain radius of DR tiers from a centroid node, and wherein NN > 2 (DR+1).

5. A method according to claim 4 comprising choosing a node as a centroid node if said node has the lowest connectivity score among nodes within DR links of said node.

6. A method according to claim 5 comprising initially choosing from a neighbourhood of candidate nodes surrounding a centroid node, a plurality of domain nodes comprising nodes within DR links of the centroid node where the centroid node has the lowest connectivity score of any centroid nodes within DR links of said candidate nodes.

7. A method according to claim 6 comprising for a node of the network: responsive to determining the node is not initially a member of a domain node and that said node has an optimal connectivity score of any nodes in a neighbourhood of said node which are not members of a domain, designating said node as a subsidiary centroid and communicating that said node has become a subsidiary centroid to nodes within a neighbourhood of said subsidiary centroid.

8. A method according to claim 7 comprising for a node of the network: responsive to receipt of an indication that a node within a neighbourhood of said node has become a subsidiary centroid, adding said node to the domain of a centroid node which is the least number of links from said node.

9. A method according to claim 6 comprising for a node of the network: responsive to determining that said node is a member of a domain comprising less than a threshold number of nodes, requesting that said node join a target domain with a centroid connected via nodes of said target domain by less than N links to said node.

10. A method according to claim 6 comprising for a node of the network: responsive to two or more centroids having the same connectivity score being the same number of links from the node, selecting the domain for the node with a centroid which has either: a) a lowest number of tiers in their neighbourhood; or

b) a lowest number of nodes in an outermost tier of their neighbourhood with an unequal number of nodes.

11. A network node operatively arranged to perform the steps of any one of claims 1 to 10.

12. A node connected to a network of nodes and including a management overlay operable to perform the steps of any one of claims 1 to 10.

13. A system comprising a plurality of network nodes according to claim 11.

14. A system according to claim 13 wherein said nodes are interconnected via either wired or wireless links.