GB2502775A - Selecting routes between nodes in a network based on node processing gain and lifetime - Google Patents

Abstract

A marginal processing gain is calculated to determine the best communication link, from a selection of candidate links, to use for a communication from a first node to a second node in a multi-node communications network. The marginal processing gain is the difference between a node processing gain with the potential link added to the node and the node processing gain without the link added. The processing gain is the difference between the total input traffic and output traffic at a node, divided by the total input traffic. The marginal processing gain is used as part of an objective function which also includes a local lifetime gain using a local lifetime defined as the minimum value out of all nodes of the residual energy at a node divided by its total energy consumption. Applications include sensor networks.

Description

Content centric and Load-balancing aware Dynamic data Aggregation

FIELD

Embodiments disclosed herein relate to the communication of data in a communications network.

BACKGROUND

Recent developments in the field of networked communications has resulted in increased demand for transport of data on mu[tihop routes, i.e. on routes, between a source node and a destination node, involving one or more intermediary nodes-The total amount of traffic to be forwarded on multi-hop routes can be significantly reduced using in-network processing and data aggregation, particularly by pre-processing of correlated information. However, a suitable data forwarding scheme is needed as a prior condition in order to efficiently compute and relay data.

BRIEF DESCRIPTiON OF THE DRAWINGS

Figure 1(a) illustrates a routing approach in the field of the described embodiment; Figure 1(b) illustrates a further routing approach in the field of the described embodiment; Figure 1(c) illustrates a yet further routing approach in the field of the described embodiment; Figure 2 is a diagram showing a network of communications nodes, illustrating an embodiment described hereLn; Figure 3(a) illustrates an example of routing in the network of figure 2; Figure 3(b) illustrates a further example of routing in the network of figure 2; Figure 4(a) illustrates an example of routing in the network of figure 2; Figure 4(b) illustrates a further example of routing in the network of figure 2; Figure 5 iLlustrates a flow diagram of a method performed by a node of the network in figure 2, in making a routing decision; Figure 6 illustrates a further example of routing in the network of figure 2; Figure 7 illustrates a graph showing performance of network lifetime with respect to variance of a weighting factor used in the abovementioned embodiment; Figure 8 illustrates a graph of performance of networks with respect to size, using an embodiment as described herein; Figure 9 illustrates energy consumption of an example of the described embodiment in comparison to prior technologies; Figure 10 illustrates a graph of coefficient of variation of node residual energy levels of an embodiment asdescribed herein, with respect to prior technologies; Figure 11 illustrates a traffic map for an embodiment described herein; Figure 12 illustrates a traffic map for an example of a prior technology; Figure 13 illustrates a residual energy map for an embodiment described herein; Figure 14 illustrates a residual energy map for an example of a prior technology; and Figure 15 illustrates apparatus capable of implementing a communications node suitable for performing a method of an embodiment described herein.

DETAILED DESCRIPTION

An embodiment provides a method of determining a communication link for a communication from a first communications node to any one of a plurality of avadable neighbouring nodes, comprising, for each available neighbouring node, determining a marginal processing gain and selecting, for the communication link, the neighbouring node with the best marginal processing gain.

The marginal processing gain may comprise a measure of communication data reduction available through aggregation, per link. The measure of communication data reduction may be normalised.

The marginal processing gain for a neighbouring node may comprise a difference between a measure of processing gain for allocation of said communication to that neighbouring node and a measure of processing gain without allocation of said communication to said neighbouring node.

The marginal processing gain may be pad of an objective function, the objective function also comprising a measure of local lifetime gain, the local lifetime gain being a function of effect on lifetime of a neighbouring node of allocation of the communication to that node.

The measure of local lifetime gain for a neighbouring node may comprise a difference between an estimate of lifetime of said neighbouring node with the allocation of said communication and an estimate of lifetime of said neighbouring node without the allocation of said communication. The measure of local lifetime gain may be normalised by the estimate of lifetime of said neighbouring node with the allocation of said communication.

The objective function may comprise a weighted sum of the measure of processing gain and the measure of local lifetime gain. The weighted sum may be dependent on a weighting factor, the weighting factor being operable to balance, in the objective function, the effect of the measure of processing gain and the measure of local lifetime gain, with respect to sensitivity to network lifetime.

The embodiment described above may comprise attaching to a communication a communication progress factor and using the communication progress factor to govern selection of candidate nodes for forward communication in the network towards an intended recipient node.

An embodiment may also comprise a communications apparatus operable in a network of communications apparatus, the apparatus being operable to determine a communication link for a communication from said apparatus to any one of a plurality of available neighbouring apparatus, the apparatus comprising a processing gain determiner operable to determine, for each available neighbouring node, a marginal processing gain and a communication link selector operable to select, as the communication link, the neighbouring apparatus with the best marginal processing gain.

Figure la illustrates a problem which may be encountered in the field, if no in-network processing takes place. In this case, communications are sent on multi-hop pathways from source nodes (indicated S1-S6) without reference to each other, to a sink node indicated Ui. As illustrated, the communications aggregate on nodes S4, S5 and SO and later hops in the communication require substantial channel capacity otherwise communication delay, failure or other deleterious consequences arise.

As a general principle, the embodiments described herein operate by taking advantage of distributed processing. A content centric and load balancing aware distributed data routing solution is presented for large-scale multi-hop M2M wireless networks.

Independent routing decisions are made by each node using only local information.

Hence, the approach is highly adaptive to dynamic environments.

A hybrid objective function for route selection is described which includes two main parts: 1. reduce the communication traffic by aggregating similar type of data, hence increasing the processing gain; 2. balance the energy-consumption among neighbouring nodes such that a longer local network lifetime can be achieved.

Eventually, in certain embodiments, the entire network lifetime can also be extended by solving the load-balancing issue on bottleneck nodes.

An embodiment provides an efficient data forwarding scheme for in-network processing to support multiple applications with heterogeneous and dynamic traffic.

An embodiment provides a hybrid method to select data relay nodes by considering communication traffic reduction, load balancing and network lifetime extension.

An embodiment provides a distributed algorithm operable to form and maintain a dynamic multi-overlaid tree topology to efficiently compute and route data in complex M2M networks.

Embodiments described herein may be implemented by any M2M multi-hop wireless network formed by mobile phones, sensor nodes, or other wireless devices.

Conventionally, in typical data gathering scenarios, information collected by nodes is first sent to a central gateway node (sink). This information is then processed for further analysis. However, in many cases, since data collected from different nodes is highly correlated, it can be combined or jointly processed while forwarding to the sink.

For example, there may be considerable correlation of data streams comprising data reports of AVERAGE or MAX readings for monitoring applications, or of data streams containing sensor readings for multiple sensors all sensing the same physical event.

In-network processing deals with this type of distributed processing of information within the network in order to reduce the total amount of messages to be sent over expensive wireless links, which has a significant impact on energy consumption as well as overall network efficiency. However, one of the main problems in this area is how data is being processed and relayed by considering various system aspects, such as multiple co-existing applications (data generated for different application may not be correlated), heterogeneous node energy levels, and load-balancing issues (some bottleneck nodes may affect the performance of the entire network due to high workload or low remaining energy), etc. Efficient data gathering and aggregation in resources constrained networks have been considered in the past. Figure lb illustrates clustering. Clustering is a simple but effective hierarchical data gathering solution, where Cluster-Heads (CH5) are polled via message gossiping among nodes in local areas and, once selected, they act as local controllers of network operations. So, as illustrated in figure Ib, node S4 has been identified as an appropriate cluster head (CH), and this gathers transmissions from nodes Si1 32, 33, S5 and $6, before transmission of an aggregated communication either through a long distance hop (indicated ©) or through a multi-hop pathway via node 55 (indicated ®). When clusters are formed, all packets in the same cluster are directly sent to the CH, and then a summary message is produced and transmitted back to the sink.

However, a periodical re-construction of the network structure (re-clustering) is required for load-balancing purposes, which incurs additional delay and extra energy consumption on communication overhead. In addition, these algorithms are vulnerable under dynamic network conditions and a homogeneous traffic pattern is usually assumed (i.e. all nodes in the network are reporting the same type of messages periodically).

Similar to clustering-based algorithms, tree-based approaches (Figure Ic) first need to construct an appropriate tree structure based on different requirements, and then the traffic flow from sdijrces to the sink (root of the tree) is routed based on the preferred directions in the tree. Data aggregation takes place once two or more messages arrive at a processing node, and their aggregate can be computed and then forwarded to the next hop.

Nonetheless, the drawback of tree-based schemes is similar to that of clustering based algorithms. Each time the traffic of an application changes or a new application arrives, the optimized tree structure need to be re-formed based on the new requirement.

Hence, in a dynamic environment with multiple applications co-existing, different data aggregation paths are required for efficient delivery of different data types and better organization of heterogeneous traffic flows. A pre-optfrnized static structure cannot satisfy this dynamic requirement. On the other hand, it is not computationally efficient to frequently reconstruct a global network topology or to compute and build multiple overlaid trees, and thus this approach would be expensive to maintain.

Embodiments will now be described which provide an efficient routing solution by integration of distributed processing and load balancing technologies for networks with dynamic and heterogeneous traffic patterns.

An embodiment will now be described from the perspective of a network, as illustrated in figure 2. The network is formed by N nodes V tvj.vq. .vd. A gateway vr sits at the centre of the network. Nodes are battery powered and each node has an finite and non-replenishment energy supply EO). Heterogeneous initial node energy levels are assumed. As in many practical systems, the gateway o is considered as a more powerful node, and is assumed to have a much higher initial energy level than other nodes, or is connected to an unlimited power supply. For simplicity, transmission power control is not enabled. Hence, all nodes have a fixed communication range and they are connected via multi-hop links. Hence, a wave-Like communication ring topology can be formed. Nodes in each ring (layer) are assigned with a layer ID representing the minimum number of hops required to reach the sink (Figure 2).

From the perspective of applications executing in the network, applications A [a.a2.a "1 randomly arrive at the gateway with a probability P and lifetime duration T Iti.t2. t3 1. For each appEication, a certain number of source nodes are required which generate the initial data. Source nodes can be pre-selected based on the application requirement (e.g. monitoring a particular area) or randomly chosen by the gateway.

Time is divided into periods called rounds, and it is assumed that traffic is generated at a homogeneous rate of r bits per packet per round for all source nodes of the same application, but different traffic rates (R [r1.r2. I) can be produced for different applications.

Once the application lifetime T falls due, corresponding source nodes will stop generating and sending data for that application. The same application can reappear in the network with probability once the previous one is terminated, and multiple different applications can co-exist in the network.

S

Figures 3a and 3b illustrate specific examples of the approach taken in this embodiment, with respect to four nodes of the network VI through V4. In the performance of this embodiment, it is assumed that data aggregation functions such as SUM, MAX, MIN, AVERAGE etc. can be performed on every node in the network (Figure 3a). In the first example, illustrated in figure 3a, the three data source nodes Vl, V2 and V3 are each running a temperature monitoring application. Hence a node V. (in this case V4) can aggregate multpIe incoming messages together with its own messages (if V1 is also a source node), into a single outgoing message.

However, it is assumed that only messages from the same application can be aggregated. As shown in Figure 3b, one of the source nodes VS is now shown running a humidity monitoring application. For a variety of reasons, different data types may not be easily processed, it may be impossible to do so in some cases. For example, it is not meaningful to calculate the average value of a data set comprising a temperature reading and a humidity reading.

In effecting the descrEbed embodiment, key to the performance Es the embedding, En each node, an objective function. Depending on a probabilEty p, the objective function is executed to rank and select the next hop node for forwarding different application traffic. Therefore, independent routing decisions are made by each node using only local information. For each data type j, the next hop node i is chosen by the objective function F which is described in equation (1): P=gg"+flL L r (1) where the first term g'-g" is the marginal processing gain which calculates the normalized communication data reduction via aggregation; L'-L the second term is defined as the normalized local lifetime gain: and

V

/3 is a tuning parameter to provide weights between the two parameters.

In the following, each term wfll be described in further depth.

Marginal processing gain: In the above expression, the marginal processing gain is given as g'-*-* g" where d is the processing gain by allocating traffic jto Node i and 2 is the processing gain without allocating trafficj to Node i; where g is calculated as g... (2) where . is the total amount of incoming traffic for all applications j relayed via node i, and is the total outgoing traffic.

Hence, the part shown on the numerator of equation (2) is the total amount of reduced traffic via the aggregation process on i and this value is then divided by the total incoming traffic.

There are two main reasons for this rationale: 1) it is a normalization process which makes it numerically comparable with the local lifetime gain (second term shown in equation (1)). Hence, a hybrid gain can be computed.

2) For load balancing purposes: it is preferred to relay traffic to a node that can provide the same processing gain (reduce the same amount of data), but with less traffic than is already assigned to it. So with the same traffic reduction amount, the more incoming traffic a node has, the smaller processing gain it can obtain.

Worked examples of the above approach are illustrated in figures 4a and 4ft In each of these examples, two apphcations (Ti & T2) are collecting data in a network formed by 6 nodes. Nodes 1, 2, 3, 4 are the source nodes of Ti and only Node 4 is providing information for application T2.

Now, Node 3 is executing the objective function to determine which node should forward its traffic.

The arrangement shown in figure 4a suggests relaying traffic from node 3 via node 5, while the arrangement shown in figure 4b is relaying traffic from node 3 via node 4.

Applying the formulations above to the situation, the following calculations arise: Scenario a Marginal processing gain of relaying traffic type TI from Node 3 to Node 5 is: = (3000-1 000)13000 = 2/3 = (4000-1000)/4000 = 3/4 Marginal processing gain = g'-g" = 1/12 Scenario b Marginal processing gain of relaying traffic type Tl from Node 3 to Node 4 is: g" (2000-2000)12000 = 0 g' = (3000-2000)/3000 = 1/3 Marginal processing gain = g'-g" = 1/3 Clearly, in this example, Node 4 (scenario b) will be selected as it has a better processing gain. This decision can be evaluated further by directly observing the traffic on each of the communication link for both cases. Although they have exactly the same amount of communication traffic on each link, Node 5 in scenario a is a bottleneck node as it needs to receive and process most of the traffic for Ti. By contrast, scenario b provides a more balanced solution.

However, if a node is equipped with more energy, in principle it can relay and process more information compared with those with less energy on board. Yet, the load balancing functionality in the processing gain function cannot reflect the heterogeneous node energy levels. Therefore, another parameter is added into the objective function set out in equation (1), known as the local lifetime gain.

Local lifetime gain: As noted above, the local lifetime gain is expressed as where LI is the local lifetime by allocating traffic ito Node i and V'is the local lifetime without allocating trafficjlo Node i; where L is calculated by: L=rnin--(3) k.N where Ek is the residual battery energy on node k; e is the total energy consumption on node k including the cost of data aggregation, read and write information in the flash, as well as for transmission and receiving data: and N is the number of candidate nodes from which the next hop node is selected.

Thus, if a Node i is the bottleneck node which has the lowest lifetime in the local region, further assigning more traffic to that node inevitably decreases the local network lifetime which also affects the overall network lifetime. Hence, in this case, a penalty is added to the objective function by the local Lifetime gain function. On the other hand, if S some messages are redirected away from the bottleneck node, a reward is given.

Hence, load balancing is achieved by not only considering the distribution of dynamic traffic flows but also heterogeneous battery energy levels Ln the network.

Rather than building a centralized overlaid tree structure for multiple applications or re-constructing each routing topology once network condition changes, a more robust way is to have a distributed decision making approach where each node decides the next hop relay based on the local information.

The operation of this embodiment is described in Figure 5. Once an application arrives at the gateway (S1-2), a pie-optimized or default routing structure is used (51-4) to collect data. However, each node i has a probability p to refine its next hop relay by executing the objective function F (S1-). A local query message is broadcast (SI-B) in the local region with its layer ID, and only its one-hop neighbouring candidates that have the same or lower layer ID will replay the query with a report message consisting of its incoming traffic and outgoing traffic information along with the node battery energy level. For each traffic type, candidate nodes are ranked by F and the one with the highest rankin is selected to relay the corresponding traffic (S 1-10).

The frequency of executing F is determined by the probability p. However, if there is no data produced on or relayed by node i, p becomes 0. Nevertheless, if new traffic appears at node i, this process continues.

Communication loops can cause many problems in multihop M2M networks such as traffic congestion, packet loss (due to Time-To-Live expiry), and additional energy consumption through the repeated processing and transmission of looping messages.

Therefore, in order to resolve this problem a reply-back constraint can be added to the local query message, such that only qualified neighbouring nodes can answer this query.

A Time-To-Go-Forward (TTGF) quantity is defined, which is an integer value representing a count to force general progress of a communication from outer layers of a network towards the intended sink. The use of this quantity in the following approach will illustrate how TTGF effects this progress.

In this embodiment, three rules are specified for the objective node to generate the query message: 1. Any parent node which has sent any application traffic to the objective node, cannot be the candidate for the corresponding traffic, unless no other qualified candidates can be found.

2. Any node with a higher layer ID compared with the objective node, cannot be selected as the candidate, unless there no other qualified candidates can be found.

3. When the Time-To-Go-Forward (TTGF) count has elapsed, only those with a lower layer ID can be the next hop candidate, unless there is no other qualified candidates can be found.

As will be understood by the reader, the use of a TTGF quantity is similar to the concept of Time-To-Live (TTL) , a bit which is added to the header of the query message. In short, if a sender has the same or lower layer ID than the recipient, the TTGF count is reduced by one. If multiple messages with different TTGF values are aggregated to a single message, the smallest TTGF value is used after the data aggregation. The TTGF value is set to default (often as a positive integer)] if a successful forward relay transmission (higher layer -. lower layer) has been made.

Worked examples of the above approach are iLlustrated in figure 6, which illustrates a portion of the network, with a first node (marked 1) in communication with various other nodes (marked 2 to 6) and thence to a base station or gateway.

Three events are identified as events (a), (b) and (c). Event (a) comprises a communication from node I to node 4, and then a decision as to which node to use for onward transmission Event (a) represents a transmission from layer 3 to layer 2. The next hop can comprise a communication to any of nodes 2, 3, 5 and 6. as indicated.

Hence, TTGF is set to default (in this case, default = 1) and all the one hop neighbouring nodes in the same or Lower layer can be the next hop candidate.

Event (b) comprises a communication from node 2 to node 4. In this case, after the TTGF value is reduced by 1 at the objective node, it is still larger than 0. Thus, neighbouring nodes in the same layer are still eligible for the next hop selection.

Hence, apart from Node 2 (which is the parent node of event (b)), nodes 3, 5 and 6 will compete to be the next hop relay.

Finally, event(c) comprises a hop from node 3 to node 4, with the TTGF being set at 1.

Since the TTGF of case (c) becomes 0 at Node 4, only those with a lower layer ID (Node 5 and 6) are qualified for candidate selection and to relay the query message.

Thus, by using TTGF, a message that, in a particular hop, has not been forwarded any "closer" to the sink, is forced to do so by selecting a lower layer node as the next hop candidate. Meanwhile, other mechanisms such as TTL can also be used such that a loop message can be discarded.

The embodiment as described herein offers the potential to improve the lifetime of the network by integration of distributed computing and load balancing technologies.

With distributed pthcessing and data aggregation, the total number of communication messages are significantly reduced, hence conserves limited energy resources. In addition, a balanced routing decision by considering dynamic traffic flows and remaining node energy levels can avoid forwarding heavy traffic to bottleneck nodes.

Therefore, a longer network lifetime can be achieved.

Independent routing decisions are made by each node using local message gossiping.

Thus, it is robust to network dynamicity and also scalable to large-scale networks.

The above described embodLment (CLDA) can be evaluated via simulation and compare it with a pre-optimized but static tree topology (STree), and the conventional centralized processing method (Central), where only the sink processes data.

The effect of the weighting factor I wiU now be discussed. Figure 7 iLlustrates a graph showing the effect of 1 on network Lifetime, for a typical network operating in accordance with the described embodiment. As can be seen, larger values of F apply more weight to the local lifetime gain in calculating the objective function, which consequently produces a smaller value of the output of the objective function for a bottleneck node. Hence] this will create a tendency for avoidance of further allocation of communications to the bottleneck node, which would otherwise shorten the local lifetime. However, as can be observed, above a certain limit in the value of 1, the impact of increasing 3 is markedly lessened.

Figure 8 shows the algorithms' performances in extending the network lifetime with different network scales. Node density is the same (0.0025 nodeftfl3). When the network width increases to 100 metres, this requires about 3 -4 hops for nodes in the farthest region to reach the sink. It can be observed that CLDA has about 30% network lifetime improvement compared with Slree and is more than four times better than the Central approach. These performance gaps become even larger when the size of the network further increases.

In Figure 9, the per round total energy consumption of processing, Read and Write data in the flash, communication, and total cost, are illustrated. The network width is set to 200 metres with the same node density. It will be noted that the communication part dominates the total energy consumption. Therefore, by taking advantage of distributed processing to reduce communication data volume] CLDA and STree save almost 75% of the energy consumption spent on communication compared with the Central method. Furthermore, the processing costs are too small to notify the differences. The costs of processing per round are listed below where CLDA and STree spent 92.1 pJ and 92.5 pJ, respectively, for distributed processing, while the Central approach has far fewer computation events involved with only half of the processing cost (42.1 pJ) compared with CLDA and STree. Nevertheless, the difference in processing is negligible in comparison with the communication part.

Figure 10 indicates the coefficient of variation of node residual energy levels. A larger value represents a higher variation in node residual energy levels which implies a poorer load balancing. It can be observed that the described CLDA has a belier performance in load balancing than both STree and Central. In addition, when simulation time passes, raw data generated from the source nodes change (e.g. arrival of a new application or termination of old ones), resulting in dynamicity in traffic flows.

As a node using CLDA can refine its route by monitoring local traffic patterns, it shows a better performance against Siree at later stages when substantial changes in S communication traffic have taken place.

In Figures 11 to 14, visualized comparisons are provided of a traffic map and a residual energy map for CLDA and Central. The line width in the traffic maps of figures 11 and 12 represent how much data is flowing via that link, while the dot size in the energy maps of figure 13 and 14 indicates corresponding node battery energy levels. It can be seen that the Central approach has much heavier communication traffic compared with CLDA. In addition, since the nodes in the area that is closer to the sink need to relay information for those located in the outer regions, a large traffic level can be observed at the centre of the network for the Central method. This could easily cause a hot-spot problem. In contrast, CLDA has much less communication data volume after aggregation. Furthermore, by observing the residual energy map, it can also be observed that CLDA conserves more energy and has a more balanced energy consumption.

While the reader will appreciate that the above embodiments are applicable to any network, and to a variety of communications apparatus in such a network, a typical apparatus is illustrated in figure 15 which provides means capable of putting an embodiment, as described herein, into effect. As illustrated, the apparatus 100 comprises a processor 120 coupled to the mass storage unit 122, and accessing a working memory 124. Although, as illustrated, user applications 126 and a communications controller 128 are represented as software products stored in working memory 124, it will be appreciated that elements of the user applications 126 and a communications controller 128 may, for convenience, be stored in the mass storage unit 122. UsuaL procedures for the loading of software into memory and the storage of data in the mass storage unit 122 apply. The processor 120 also accesses, via bus 130, a user input unit 136 and a user output unit 138. A communications unit 132 operates to effect communications, either wireless or wired, with other apparatus.

Execution of the communications controller software 128 by the processor 120 causes an embodiment as described herein to be implemented. The communications controller software 128 can be embedded in original equipment, or can be provided, as a whole or in part, after manufacture. For instance, the communications controller software 128 can be introduced, as a whole, as a computer program product, which may be in the form of a download, or to be introduced via a computer program storage medium, such as an optical disk. Alternatively, modifications to an existing communications controller 128 can be made by an update, or plug-in, to provide features of the above described embodiment.

Embodiments described herein can conceivably be implemented in any of a wide range of wireless networks for multi-point to point routing purposes, such as wireless sensor networks, ad-hoc networks, Wi-Fl mesh etc. Particularly, for data collection in resource constrained M2M networks, a large number of heterogeneous sensor nodes are employed for continuous sensing and data gathering. An efficient data aggregation and delivery scheme can significantly extend the network lifetime. Hence, embodiments as described herein offer a potential for significant savings on network maintenance and to cut down node redeployment cost.

While certain embodiments have been described, these embodiments have been presented by way of example only, and are not intended to limit the scope of the inventions. Indeed, the novel methods and apparatus described herein may be embodied in a variety of other forms; furthermore, various omissions, substitutions and changes in the fom of the methods and apparatus described herein may be made without departing from the spirit of the inventions. The accompanying claims and their equivalents are intended to cover such forms or modifications as would fall within the scope and spirit of the inventions.

Claims (12)

1. CLAIMS: 1. A method of determining a communication link for a communication from a first communications node to any one of a plurality of available neighbouring nodes, comprising, for each available neighbouring node, determining a marginal processing gain and selecting, for the communication link, the neighbouring node with the best marginal processing gain.
2. 2. A method in accordance with claim 1 wherein the marginal processing gain comprises a measure of communication data reduction available through aggregation, per link.
3. 3. A method in accordance with claim 2 wherein the measure of communication data reduction is normalised.
4. 4. A method in accordance with any one of the preceding claims wherein the marginal processing gain for a neighbouring node comprises a difference between a measure of processing gain for aLlocation of said communication to that neighbouring node and a measure of processing gain without allocation of said communication to said neighbouring node.
5. 5. A method in accordance with any one of the preceding claims wherein the marginal processing gain is part of an objective function, the objective function also comprising a measure of local lifetime gain, the local lifetime gain being a function of effect on lifetime of a neighbouring node of allocation of the communication to that node.
6. 6. A method in accordance with claim 5 wherein said measure of local lifetime gain for a neighbouring node comprises a difference between an estimate of lifetime of said neighbouring node with the allocation of said communication and an estimate of lifetime of said neighbouring node without the allocation of said communication.
7. 7. A method in accordance with claim 6 wherein said measure of local lifetime gain is normalised by the estimate of Lifetime of said neighbouring node with the allocation of said communication.
8. 8. A method in accordance with any one of claims 5 to 7 wherein the objective function comprises a weighted sum of the measure of processing gain and the measure of local lifetime gain.
9. 9. A method in accordance with claim 8 wherein the weighted sum is dependent on a weighting factor, the weighting factor being operable to balance, in the objective function, the effect of the measure of processing gain and the measure of local lifetime gain, with respect to sensitivity to network lifetime.
10. 10. A method in accordance with any preceding claim including attaching to a communication a communication progress factor and using the communication progress factor to govern selection of candidate nodes for forward communication in the network towards an intended recipient node.
11. 11. A communications apparatus operable in a network of communications apparatus, the apparatus being operable to determine a communication link for a communication from said apparatus to any one of a plurality of available neighbouring apparatus, the apparatus comprising a processing gain determiner operable to determine, for each available neighbouring node, a marginal processing gain and a communication link selector operable to select, as the communication link, the neighbouring apparatus with the best marginal processing gain.
12. 12. A computer program product comprising computer executable instructions which, wl-ien executed by a computerised communications apparatus, causes that apparatus to perform a method in accordance with any one of claims ito 10.