Method for optimizing utilization distribution in a communication network - Patents.com

Abstract

The present invention relates to a computer-implemented method for optimizing utilization fraction routing in a communication network (1) using a quantum conceptual processor (6). A set of traffic requests for transfer of a determined data volume between an origin node o and a destination node d of a plurality of communication nodes 2 is captured. The traffic requests (5) are split into sub-requests (7, p). A set of alternative communication paths (k) for individual routing of each sub-request (7, p) is specified. Edges (e) in the set of alternative communication paths (k) are assigned respective utilization capacity limits. The fractional capacity utilization of the edges (e) is calculated based on the respective utilization capacity limits. The calculated fractional capacity utilization is then formulated as a term of a quadratic stress function. Using the quantum conceptual processor 6, an optimized routing is determined by selecting one communication path (k) from the set of alternative communication paths (k) for each sub-request (7, p) such that the quadratic stress function is minimized.

Description

[Technical Field]

[0001] The present invention relates to a computer-implemented method for optimizing utilization distribution in a communication network through which data traffic is routed, the communication network having a plurality of communication nodes connectable via communication path edges (links) for routing data traffic. The present invention also relates to a quantum conceptual processor configured to perform such a method, and a computer program implemented to perform such a method. [Background technology]

[0002] Today's demands for data traffic in communication networks are increasing dramatically in these times. With the recent introduction of 5G, an increasing number of devices and applications are pushing data traffic to new peaks. Furthermore, the increasing demands of digitalized and distributed work and the increasing demands for streaming in private household domestic environments are also contributing significantly to this trend. The increasing amount of data transferred over communication networks such as the Internet poses a major challenge to service providers. To avoid communication network congestion and degradation of user experience, traffic engineering techniques can be deployed to complement the relatively slow and expensive expansion of network infrastructure.

[0003] The most widely deployed engineering techniques for data traffic management in communication networks operate on the premise that communication paths are calculated in terms of link weights. These weights are often related to link capacity, i.e., the maximum amount of data per unit time that can be routed through a link from a starting node to its respective edge-connected end nodes. The final routing of a data stream from its origin node to its destination node is based on the identified path of the shortest path found, taking into account intermediate nodes and a given link or edge. As a result, the simplest and most practical technique for guiding traffic demand is through the manipulation of these link weights, also known as link metrics or IGP (Interior Gateway Protocol) metrics. The higher the link weight of an edge, the higher the probability that data will be routed through the respective edge. Following this approach, link weights are locally adapted in a reactive manner whenever a particular link tends to become overloaded. In a more systematic manner, this problem is further addressed by applying linear integer computer programs, where the optimization goal is to minimize the maximum link capacity usage in the network. However, the task of finding a globally optimal set of metrics is extremely complex. In terms of computational complexity, this task is NP-hard because each link metric can affect many communication paths.

[0004] Linear optimization techniques applied so far quickly reach their limits when considering real-world nonlinear conditions such as redundancy, geographical subgroups or subdomains (e.g., European and US networks considered in one model), satellite inclusion, quality of service (QoS), relationships, etc. Furthermore, known techniques often lead to problems of unused capacity utilization and link capacity overload in communication paths within the network, where many links are close to their capacity limits. Summary of the Invention

[0005] It is therefore an object of the present disclosure to provide an enhanced technique that allows optimized utilization of communication paths in a communication network with respect to capacity limitations, thereby achieving optimized routing.

[0006] This problem is solved by the method according to claim 1. Further embodiments are set out in the dependent claims and the following description.

[0007] The method is a computer-implemented procedure for optimizing utilization fraction in a communication network through which data traffic is routed, the communication network having a plurality of communication nodes. The communication nodes are connected by edges of the communication network. A set of edges generates a communication path for routing the data traffic. Thus, an edge of a communication path in this context describes a connection between two adjacent nodes in the communication path.

[0008] The method comprises: - capturing a set of traffic requests, each traffic request specifying a transfer of a determined data volume from an origin node to a destination node of a plurality of communication nodes; - splitting the traffic request into sub-requests; - designating a set of alternative communication paths for individual routing of each sub-request, where edges in the set of alternative communication paths are assigned respective utilization capacity limits; - for each sub-request, calculating a fractional capacity utilization of edges in the set of selective communication paths, the fractional capacity utilization being calculated based on a respective utilization capacity limit; - formulating the calculated fractional capacity utilization as a term of a quadratic stress function; - determining an optimized routing by selecting, for each sub-request, one communication path from the alternative communication paths such that a quadratic stress function is minimized, using a quantum conceptual processor.

[0009] The method reliably addresses the problem of routing network demands in a communication network along optimized communication paths such that the overall capacity in the network is optimally used, thereby avoiding link capacities in the network being exceeded.

[0010] By applying this method, for each given sub-request from a traffic request, one optimal choice for a communication path can be selected from a set of alternative communication paths. This selection is made such that the capacity of all edges (links) in the used communication paths in the network is respected as an upper bound on the total volume of traffic requests routed along them, and the average load of all communication paths in the network is minimized. Furthermore, minimization of maximum link utilization (MLU) can be achieved.

[0011] A "traffic request" in this context is modeled as a 3-tuple, defining an origin node (the source of the data stream), an end node or destination node (the destination of the data stream), and the determined data traffic to be transferred between the origin and destination. The focus is on providing a continuous data stream over the network, which is modeled and routed in such a way that data is not lost during transmission by exceeding the specified capacity on a given transport link. Measurement of the data transfer rate of such data stream requests or requests is currently specified in Gbps (Gigabits per second).

[0012] A "sub-request" in this context is a traffic request that has been split into fragments. Thus, one sub-request represents a fragment of an initial traffic request with respect to a determined data volume split into data volume packets.

[0013] The selective communication paths in this context are generally not subject to any restrictions regarding routing, path length, or the number of intermediate nodes in the network. However, a set of selective communication paths is predetermined for each subrequest transmitted through the network. In such a predetermined determination, useful or preferred paths may be considered in terms of latency (shortest possible path, fewest possible IP hops), redundancy (the model should be redundant against connection failures), domain (e.g., EU, US) or hierarchy (core network, access network), etc. For example, the set of selective communication paths is a subset of possible communication paths for each traffic request or each respective subrequest. The set of selective communication paths is stored, for example, as a "route box" that can be accessed by a computer-implemented algorithm. Advantageously, an appropriate number of divergent (most diverse or disjoint) paths are preselected in the route box to provide a sufficiently large solution space for solving the quadratic stress function, i.e., finding a (global) minimum, by a quantum conceptual processor. Such a preselection may depend on the processing performance and capacity of the quantum conceptual processor.

[0014] Furthermore, a traffic request in this context can theoretically be split into sub-requests with any equal or unequal fragment size that is suitable for practical implementation. The approach here is to split each traffic request into multiple sub-requests and find an optimal communication path through the network for each sub-request. In this way, this approach is based on so-called MCFR (Multi Commodity Flow Routing), a type of source routing. Such traffic request splitting may depend on the processing performance and capacity of the quantum computing processor.

[0015] A quadratic optimization problem can be formulated to address the complexity of the optimization problem described above by calculating, for each sub-request, the fractional capacity utilization of the edges in the set of alternative communication paths and formulating the calculated fractional capacity utilization as terms of a quadratic stress function. Applying such a quadratic optimization problem has the advantage that a quadratic stress function can be formulated that strongly penalizes high capacity utilization on individual edges of the communication paths.

[0016] In this way, an optimized routing is determined by selecting, for each sub-request, one communication path from the set of alternative communication paths such that the quadratic stress function is minimized. The minimum of the quadratic stress function is preferably a global minimum, but can also be a local minimum.

[0017] Therefore, the method has the technical effect and advantage of uniform minimum utilization of the network and distribution of distances to capacity limits within the network to achieve uniform minimum utilization of the network with respect to capacity limits.

[0018] The underlying quadratic optimization problem, as described above, is highly complex. This is not only due to the potential impact of one selected communication path on other paths and the massive amount of data traffic managed between multiple origin and destination nodes in the network. It is also due to the many practical constraints that must be considered. As more constraints are implemented, such problems become more complex and difficult to solve. This can be problematic or challenging when a traffic engineering solution is needed quickly, for example, in response to unexpected network failures, or when additional practical constraints are considered, such as latency (shortest possible path, fewest possible IP hops), redundancy (the model should be redundant against one or more / many edge failures, planned outages, or network link maintenance), domain (EU, US) or tier (core network, access network). The method described herein advantageously demonstrates its strengths compared to conventional approaches where the underlying problem becomes increasingly complex. In other words, for complex optimization problems that consider practical constraints such as those described above, the method described herein has significant advantages over conventional techniques.

[0019] The method described herein utilizes a quantum computing-inspired approach. The calculation of the optimal solution of a quadratic stress function to determine an optimized communication path for all subrequests of a set of traffic requests is performed by a so-called quantum processor. A quantum processor in the context of this disclosure is defined as a processor that solves the so-called "Ising model" or an equivalent quadratic unconstrained binary problem. For example, it is a processor configured to solve optimization problems using quantum annealing or quantum annealing emulation. Such processors are based on, for example, conventional hardware technologies, such as complementary metal-oxide semiconductor (CMOS) technology. An example of such a quantum processor is Fujitsu's Digital Annealer. Alternatively, any other quantum processor can be used for the method described herein, and in the future, technologies based on actual qubit technology may also be used. Further examples of such quantum processors are DWave's Quantum Annealer (e.g., 5000Q), as well as quantum gate computers (IBM, Rigetti, OpenSuperQ, IonQ, or Honeywell) that utilize quantum optimization algorithms such as QAOA or VQE.

[0020] In other words, a quantum conceptual processor as defined herein is a processor that implements the concept of minimizing so-called quadratic unconstrained binary optimization (QUBO) functions, either in classical technology in specialized processors, quantum gate computers, or quantum annealers.

[0021] In at least one implementation, the method includes: - specifying a set of path variables, each path variable being associated with one of the sub-requests and one communication path from a set of alternative communication paths; - formulating a path term in a quadratic stress function, the path term connecting the calculated fractional capacity utilization of an edge of each communication path from the set of alternative communication paths with a path variable associated with each communication path from the set of alternative communication paths; - using the quantum conceptual processor to calculate, for each subrequest, a path term for selecting one communication path from the set of alternative communication paths such that a quadratic stress function is minimized.

[0022] In this way, for each sub-request, an optimal routing between the origin node and the destination node on one selected communication path along the concatenation of connections between adjacent nodes in the network can be calculated separately. This provides an elegant implementation of highly flexible and variable routing of data traffic, especially when considering the MCFR approach as described above. Thus, different communication paths for different sub-requests (e.g., via different intermediate nodes) can be selected to avoid overloading or significant increases in capacity utilization at each edge of the communication path in the network and to distribute the overall capacity utilization in an optimized manner throughout the network.

[0023] By connecting the calculated partial capacity utilization of edges in the selected communication paths with the path variables associated with each communication path, an optimal solution (minimum) of the quadratic stress function for all sub-requests can be calculated. In this way, optimal selection of one path from the path box for each sub-request can be achieved to satisfy the optimization problem described above. Therefore, the impact of the communication path selected for one sub-request on other possible communication paths for other sub-requests can be mitigated. This allows for a great deal of routing freedom, but is very complex to solve. The optimized selection of each path from the path box for all sub-requests is performed by a quantum conceptual processor, as described above.

[0024] In at least one implementation of the method, the path terms are calculated taking into account the path condition that each sub-demand is routed along exactly one communication path from a set of alternative communication paths. Such path conditions form constraints or "boundaries" on the method so that each sub-demand is assigned to exactly one path from the path box. This avoids undesirable solutions and ensures that the routing of each sub-demand is fully considered.

[0025] In at least one implementation of this method, a traffic request is split into subrequests having determined discrete data volumes. The subrequests can each have the same size or different sizes, depending on the implementation and practical considerations. For example, a traffic request with a volume size of 1000 Gbit / s is split into multiple subrequests of even size, 50 Gbit / s. Alternatively, subrequests having different sizes may be generated, with the different subrequests having different sizes, e.g., 50, 100, and 250 Gbit / s. Splitting a traffic request into subrequests having determined discrete data volumes has the effect of an algorithmic procedure that can be practically implemented in the network and helps maintain stable and reliable control of the data stream. In this way, such an approach is a type of discrete MCFR approach.

[0026] In at least one implementation of the method, the quadratic stress function is based on the following constraints on the set of traffic demands or each sub-demand: - organization of communication networks in different network domains; - The latency of the communication network is taken into consideration.

[0027] By taking such constraints into account in the formulation of the quadratic stress function, solutions to the optimization problem that violate the above conditions can be penalized, thereby making it possible to find a suitable optimal solution that takes into account the practical constraints of actual network conditions in a communication network.

[0028] In at least one implementation of the method, the set of alternative communication paths for the individual routing of each sub-request is determined according to the following constraints: - one or more redundant selective communication paths associated with a sub-network of the communication network (1); - organization of communication networks in different network domains; - the latency of the communication network.

[0029] This has the advantageous effect that the optimized routing calculation can react to and compensate for failures in zones, segments, or sub-networks within the communication network, take into account different domains of the network, and / or react to and compensate for latency in the network. This also provides an additional degree of freedom, allowing each sub-network, domain, and latency to be emphasized in the optimized routing calculation for every sub-request. For example, some zones or regions within a communication network may have greater importance, significance, or usage density than other zones or regions. This can be countered by such measures. Also, a communication network can be segmented into different sub-networks to better handle different latency requirements in this regard.

[0030] In at least one implementation of the method, the set of alternative communication paths for the individual routing of each subrequest is specified such that fewer alternative communication paths are selected for topologically close origin and destination nodes than for topologically distant origin and destination nodes. This has the advantage that all possible combinations and options of communication paths can be condensed into a suitable number of alternative paths in the path box for each respective subrequest. For topologically close origin and destination nodes, fewer alternative communication paths are sufficient, while for topologically distant origin and destination nodes, a greater number of alternative communication paths is recommended. For close origin and destination nodes, rather short paths are preferred, while for distant origin and destination nodes, sufficient alternative routes or detours can be considered. Thus, as the "distance" between the origin and destination nodes increases, suitable and sufficient options and alternatives can be predetermined as alternative communication paths in each case without excessively burdening the complexity of the algorithm.

[0031] In at least one implementation of this method, the quadratic stress function is formulated as a quadratic unconstrained binary optimal (QUBO) function. This QUBO function serves as the "input" of a quantum conceptual processor that solves this optimization problem for optimized routing of all subrequests in the manner described above. Generally speaking, a QUBO is a quadratic polynomial of binary variables represented in the quantum conceptual processor as bits or quantum bits (hereinafter, Qbits). In the context of the optimization problem of this disclosure, the QUBO function represents the sum of the possible contributions of the partial capacity utilization of each edge in a selective communication path as a function of different Qbits, where each Qbit represents a selection of path alternatives that can assume the value "0" or "1." To solve the quadratic optimization problem (quadratic stress function), the quantum conceptual processor performs different settings of different Qbits to find a solution that minimizes the quadratic optimization problem. In this way, the QUBO representation of the optimization problem has elegant properties with respect to quantum conceptual computing as applied herein. For example, the path variables described above are formulated in the form of such Qbits.

[0032] In at least one implementation of the method, the quadratic stress function and the path conditions are weighted and combined into a global QUBO function, as described above. The global QUBO function can take into account one or more of the constraints described above. In this regard, one or more of the constraints described above can be weighted within the QUBO function as soft constraints. This has the advantage that the QUBO function can be fine-tuned to some extent depending on the focus of the optimization problem: either optimizing a uniform capacity utilization distribution across the entire network, or achieving one or more of the aforementioned (soft) constraints.

[0033] The above-mentioned problem is also solved by a quantum conceptual processor as claimed in the accompanying claims. The quantum conceptual processor is configured to perform one or more steps of the method as described above. According to an exemplary implementation, the quantum conceptual processor is a digital annealing processing unit. This unit may be specially configured to perform quantum annealing or quantum annealing emulation as described above. The quantum conceptual processor may be of any type described above.

[0034] The above problem is also solved by a computer program comprising instructions, which when executed by one or more processors, cause each of the one or more processors to perform one or more steps of the above-described method. At least one of these processors may be, for example, a quantum conceptual processor as described above. Other processors may be configured to process preliminary or repetitive steps of the above-described method by executing the computer program.

[0035] The above-mentioned problem is also solved by a workplace for a network planner to verify an optimized route determined by the above-described method. Such a workplace has, for example, a verification means configured for (automatic or semi-automatic) verification of an optimized route determined by the above-described method. This serves the network planner to verify the optimization result found by the above-described method. The verification means can be implemented in software and / or hardware. For example, the workplace can be in communication with or connected to a system including a quantum computing processor that executes the above-described method. The results can then be passed on to the workplace.

[0036] Furthermore, the above-mentioned problem is also solved by an interface configuration including one or more interfaces to a plurality of communication nodes of a communication network through which data traffic is routed, the interface configuration being configured to automatically deploy the optimized routing determined by the above-described method to the communication nodes of the communication network. In this way, the optimized routing determined by the above-described method can be deployed (automatically or semi-automatically) to the plurality of communication nodes of the respective communication network. For example, the interface configuration can be in communication with or connected to a workplace as described above or a system including a quantum computing processor that executes the above-described method. The results can be passed on to the interface configuration.

[0037] Additionally, as a preliminary measure for one or more of the above-described steps of the computer-implemented procedure, an interface may be implemented or used to read parameters from the communication network prior to each optimization and input such parameters into the described computer-implemented optimization procedure, including, for example, the network configuration, neighbor information for a graph description of the network, available capacity within the network, and expected traffic demands.

[0038] Any aspect, feature, advantage, or measure described in connection with the method described above, either alone or in combination with each other, may apply to, or find similar expression in, any aspect, feature, advantage, or measure described in connection with the quantum conceptual processor or computer program described above, either alone or in combination with each other, and vice versa. [Brief explanation of the drawings]

[0039] The invention is further described below with the aid of several drawings and in view of several implementations.

[0040] [Figure 1]1 illustrates an exemplary configuration of a communications network with exemplary routing of traffic demands according to a conventional approach. [Figure 2A] 1 illustrates an exemplary configuration of a communication network having exemplary routing of traffic demands according to an alternative approach. [Figure 2B] 1 illustrates an exemplary configuration of a communications network with exemplary routing of traffic demands in accordance with an approach according to the present invention; [Figure 3] 1 illustrates an exemplary schematic diagram of selective communication paths for routing traffic requests between an origin node and a destination node. [Figure 4A] 1 illustrates an exemplary mathematical formulation of a partial optimization problem according to the inventive approach. [Figure 4B] 1 illustrates an exemplary mathematical formulation of a partial optimization problem according to the inventive approach. [Figure 5] 1 shows an exemplary schematic diagram of an algorithm implementing the approach according to the invention;

[0041] 1 shows an exemplary configuration of a communication network 1 with exemplary routing of traffic requests 5a, 5b, and 5c according to a conventional approach. The communication network 1 includes a plurality of communication nodes 2, with a connection 4 between two adjacent communication nodes 2 referred to as an edge. This is shown between communication node 2 and another communication node 2c, which can communicate with each other via the connection 4. Depending on the historically grown configuration and implementation of the communication network 1, some communication nodes 2 are aggregated into so-called aggregation nodes 3. As exemplarily shown in FIG. 1, for example, communication node 2a is aggregated into aggregation node 3a, and other communication nodes 2b, 2d, and 2e are aggregated into aggregation node 3b.

[0042] The communication nodes 2 are, for example, so-called label edge routers (LERs) for routing incoming and outgoing data traffic within the network 1. The aggregation nodes 3 are called metanodes and are aggregation zones of LERs in a particular area of ​​the network 1. For example, the aggregation nodes 3 are central aggregation zones of a determined economic region or city where communication should take place. In other applications, the aggregation nodes 3 can be entities such as, for example, industrial networks or traffic networks.

[0043] The communication network 1 is generally partially mashed. This means that not all communication nodes 2 are connected to all other communication nodes 2. Instead, there are only a few connections 4 (see dotted connections) between some communication nodes 2 implemented in the network 1, resulting, for example, from the historical development of the network 1. The connections 4 between the respective communication nodes 2 are implemented, for example, by optical fiber connections. However, other technologies, such as wireless technologies (e.g., 5G) and copper / DSL technologies, are generally also applicable.

[0044] As explained above, FIG. 1 illustrates a specific scenario of traffic requests 5a, 5b, and 5c, according to which a specific amount of data must be transferred between respective communication nodes 2 in network 1. As illustrated exemplarily, a first traffic request 5a is between a communication node 2a in aggregation node 3a and another communication node 2f in aggregation node 3d. A second traffic request 5b is between a communication node 2e in aggregation node 3b and a communication node 2f, again in aggregation node 3d. A third traffic request 5c is between a communication node 2c in aggregation node 3c and a communication node 2f, again in aggregation node 3d. Thus, each traffic request 5a, 5b, and 5c defines a determined amount to be transferred from an originating node to a destination node. In the exemplary scenario according to FIG. 1, the originating node for traffic request 5a is communication node 2a, while the destination node for traffic request 5a is communication node 2f. Similarly, for traffic request 5b, the originating node is communication node 2e and the destination node is communication node 2f. Similarly, for traffic request 5c, the source node is communication node 2c and again the destination node is communication node 2f.

[0045] In an alternative implementation, traffic demands can be defined as demands between aggregation nodes 3, regardless of which internal communication node 2 within each aggregation node 3 the communication originates or terminates at. For example, demands 5a, 5b, and 5b can be defined as the demand between aggregation nodes 3a and 3d (demand 5a), the demand between aggregation nodes 3b and 3d (demand 5b), and the demand between aggregation nodes 3d and 3d (demand 5c). In such an implementation, there is a "virtual" edge between each aggregation node and its internal communication node, and the virtual edge has a very high capacity. This leads to the effect that where communication originates or terminates at an internal communication node 2 within each aggregation node 3 does not play a significant role.

[0046] Each traffic request 5a, 5b, and 5c burdens the network 1 with a utilization of the network capacity, i.e., the utilization of the capacity of each connection 4 of the possible communication paths between each communication node 2 in the network 1. In the exemplary scenario of FIG. 1, traffic request 5a is forwarded from communication node 2a to communication node 2f via communication nodes 2b, 2c, 2d, 2e, and 2f. In parallel with this, traffic request 5b is simply forwarded via connection 4 between communication nodes 2e and 2f. Furthermore, traffic request 5c is forwarded from communication node 2c to communication node 2f via communication nodes 2d, 2e, and 2f. In this scenario, two main drawbacks arise. The first drawback is that the communication paths for forwarding traffic requests 5a and 5c are long and involve rather complex routes through the network 1. These forwardings involve multiple communication nodes 2 and connections 4 in the network 1 to forward traffic requests 5a and 5c, which may significantly impact other forwardings in the network. A second drawback lies in the fact that all traffic requests 5a-5c are ultimately forwarded via the connection 4 between the communication nodes 2e and 2f. Therefore, the link capacity of the connection 4 between the nodes 2e and 2f is strained to a considerable extent. This may lead to overloading of the connection 4 between the nodes 2e and 2f, resulting in increased latency or loss of data, etc.

[0047] 2A shows an exemplary configuration of communication network 1 with exemplary routing of traffic requests 5a, 5b, and 5c (see above) according to an alternative approach. In the scenario according to FIG. 2A, an alternative path is selected for traffic request 5a such that traffic request 5a is forwarded over an alternative communication path that again starts at communication node 2A and continues to communication nodes 2b, 2d, 2e, and 2f. The alternative path is selected, for example, by manipulating link weights of connections 4 within network 1.

[0048] The scenario according to Fig. 2A is advantageous over the scenario of Fig. 1 in that the communication path for the forwarding of traffic request 5a is closer to a short path strategy, thereby keeping the number of involved communication nodes 2 and connections 4 in network 1 low (at least lower than in the scenario of Fig. 1) in order to reduce the impact on other forwardings. However, even in the scenario of Fig. 2A, another drawback remains, according to which connection 4 between nodes 2e and 2f is still heavily loaded as all three of traffic requests 5a-5c pass through this connection in network 1.

[0049] FIG. 2B illustrates an exemplary configuration of the communication network 1 according to FIGS. 1 and 2A, but in which exemplary routing of traffic requests 5a-5c follows the approach of the present invention. In the scenario of FIG. 2B, multiple communication paths are selected that are optimized for the data volumes of all three traffic requests 5a-5c. In this regard, traffic requests 5a-5c are all split into multiple subrequests, each subrequest representing a discrete fragment of the data volume of the associated one of traffic requests 5a-5c. This is illustrated in FIG. 2B, where traffic requests 5a-5c are indicated only by dashed arrows. Thus, according to the implementation of FIG. 2B, there is not just one defined path for each traffic request 5a-5c, but multiple (different, separate) paths for each subrequest of each traffic request 5a-5c.

[0050] Thus, the scenario according to Figure 2B transfers the overall data volume of traffic requests 5a-5c split into multiple fragments (sub-requests) over very different communication paths through the network from their respective origins to their respective destinations. This approach therefore follows a discrete MCFR approach. The real value of the scenario according to Figure 2B lies in the fact that the overall data volume of all traffic requests 5a-5c is distributed across the entire network, thereby avoiding transferring a large load across a single segment in the network and placing a large burden on a single connection 4 in the network.

[0051] Therefore, the scenario of Figure 2B overcomes the drawbacks of the approach according to Figures 1 and 2A, thereby achieving a uniform and optimized distribution of the overall capacity utilization of the connections 4 in the network 1 for all traffic requests 5 that have to be forwarded within the network 1, while reducing the influence and adoption between the forwarding of different traffic requests 5a to 5c as much as possible.

[0052] In the following, the implementation of the approach according to FIG. 2B is explained in more detail.

[0053] The optimization problem to be solved is to determine an optimized routing through the network 1 by selecting one communication path from a set of alternative communication paths for each sub-request as a fragment of each traffic request 5 such that a mathematically formulated quadratic stress function (core optimization problem) is minimized. This serves the purpose of selecting a respective communication path for all sub-requests of all traffic requests 5 in the network 1 with the effect that the overall capacity utilization of the connections 4 in the network 1 can be minimized uniformly across the network 1. This avoids placing a heavy or excessive load on any connection 4, while the light load on other connections 4 can significantly reduce such stress.

[0054] To achieve the above effects, a computer-implemented algorithmic method for optimizing routing within the communication network 1 is implemented, which is described below.

[0055] 3 shows an exemplary schematic diagram of selective communication paths k1 and k2 for routing fragments (sub-requests) of a traffic request between an origin node o and a destination node d. Multiple sub-requests of a determined data volume (into which traffic request o, d is split) are transferred between o and d. Here, the core optimization problem is to select and determine optimally distributed communication paths for all sub-requests between o and d such that the overall capacity utilization of connections in the selective communication paths is minimized, such that the overall utilization of capacity in the network is minimized uniformly in the network.

[0056] According to FIG. 3, a set of alternative communication paths k1-k2 is pre-specified. This can be done through the application of any suitable path planning algorithm that calculates an alternative communication path between an origin node o and a destination node d for each sub-request to be forwarded. As exemplarily shown in FIG. 3, path k1 runs from o1 to d1 via an intermediate node i1. Path k2 runs from o1 to d1 via another intermediate node i2. These are alternative communication paths for routing data traffic included in each sub-request from the origin o to the destination d. The planned paths k1 and k2 are pre-stored in the form of a path box and can be accessed by an algorithm for selecting one path for each sub-request from the path box.

[0057] 3 further illustrates two exemplary connections within optional communication paths k1, K2, further referenced as edges e1 and e2: edge e1 is configured between intermediate node i1 and destination d, and edge e2 is configured between intermediate node i2 and destination d.

[0058] Let us exemplarily assume different options for routing two sub-requests between origin o and destination d based on route boxes k1 and k2. One option for per-sub-request routing is route k1, where data traffic is forwarded via edge e1. Another option for per-sub-request routing is route k2, where data traffic is forwarded via edge e2. As can be seen from these different options for routing data traffic per sub-request, there are combinations of communication routes for each of the two sub-requests, where two edges e1 and e2 are each loaded with only one sub-request. This is given, for example, by one sub-request passing through route k1 and the other sub-request passing through route k2. However, there are also possible combinations of communication routes where one of edges e1 and e2 is substantially and heavily loaded with both sub-requests, while the other edge e1 and e2 is not used at all. This is given by both sub-requests passing through the same route k1 or k2.

[0059] The latter combination has the significant drawback that the capacity utilization of one of edges e1 and e2 may be significantly higher, potentially resulting in overload or failure of the respective edge. Therefore, the optimization problem consists in determining and selecting distributed communication paths for all sub-demands between o and d such that the overall capacity utilization is distributed across both edges e1 and e2.

[0060] To solve this optimization problem, the fractional capacity utilization of all edges in the set of alternative communication paths k1, k2 is calculated for the entire set of sub-demands into which all traffic demands are split. As exemplarily given in Figure 3, such a strategy involves calculating the fractional capacity utilization of each of edges e1 and e2 for each of the sub-demands between o and d. The "fractional capacity utilization" of each edge means that the fraction of capacity utilization required for each sub-demand transmitted over this edge is calculated based on each edge's respective utilization capacity limit.

[0061] For example, with reference to FIG. 3, assume that each sub-demand between o and d requires half of the maximum utilization capacity of each edge e1 and e2 when traversing the respective edge (i.e., 50% of capacity). This means that edges e1 and e2 are burdened with half of their utilization capacity for each sub-demand between o and d when each sub-demand traverses the respective edge. In other words, if one sub-demand traverses path k1 and the other sub-demand traverses path k2, then both edges e1 and e2 are burdened with 50% of their utilization capacity limits. Otherwise, if both sub-demands traverse the same path k1 or k2, then each edge e1 (for k1) or e2 (for k2) will be fully and completely burdened (2 x 50% = 100%), thereby reaching its capacity limit and resulting in utilization of the entire edge capacity.

[0062] Such a calculation of fractional capacity utilization is performed for all remaining edges within alternative communication paths k1 and k2 in the scenario of Figure 3. The calculated fractional capacity utilization is then formulated in terms of a quadratic stress function, as described in more detail below and with reference to Figure 4B.

[0063] Figures 4A and 4B show an exemplary mathematical formulation of a partial optimization problem following the approach as described above with respect to Figures 2B and 3. The mathematical formulation in Figures 4A and 4B is expressed as a so-called Hamiltonian function, a short Hamiltonian.

[0064] The mathematical formulation in Figure 4A formulates the path condition by saying that each sub-request between each origin and destination (o, d) is routed along exactly one path k from a path box P (k∈P). For the scenario in Figure 3, this means that each sub-request between o and d goes through either k1 or k2.

[0065] The mathematical formulation of FIG. 4A illustrates a binary variable that can assume the value “0” or the value “1” (or both with a particular probability) and is represented in a quantum-mechanical processor as a bit (or q-bit as used below).

number

number

number

number

number

[0066] The mathematical formulation of the Hamiltonian according to Figure 4B represents the core optimization problem formulated as a quadratic stress function that takes into account the calculated partial capacity utilization of edge e (e∈E) among all edges E in the network, where edge e is in the set of alternative communication paths k∈P for all traffic demands o, d. Thus, the core optimization problem here is to minimize the Hamiltonian according to Figure 4B in order to find optimized communication paths for all sub-demands transmitted in the network.

[0067] Assuming that the equation in Figure 4A is satisfied, the Hamiltonian in Figure 4B considers a summation term for each edge e in the selective communication path k, taking into account all subrequests p between all origins o and destinations d. Thus, the QUBO in Figure 4B then calculates all the calculated fractional capacity utilizations of all edges e that are part of the selective communication path k.

number

number

number

number

number

number

number

[0068] Considering the scenario of FIG. 3 for the example of two sub-requests p1, p2 and two edges e1 and e2, the Hamiltonian according to FIG. 4B may have the following formula:

number

[0069] Under the assumption that each sub-demand p1, p2 burdens each edge e1, e2 with half (50%) of its capacity, the above term reaches a minimum when p1 and p2 are routed via different paths k1, k2, as explained above. Then, the above term is as follows:

number

[0070] Otherwise, if p1 and p2 are routed via one common path k1 or k2 (no other paths are used), then the above term is:

number

[0071] Therefore, the network cost / stress is higher in the latter solution and worse than the above solution.

[0072] The above example shows that selecting different paths k1, k2 for two sub-requests p1, p2 is a preferred solution to achieve the optimization goal of overall capacity utilization distribution across the network.

[0073] The Hamiltonian in Figure 4B generally represents the number of Qbits for all subrequests in the network.

number

number

number

number

[0074] Fig. 5 shows an exemplary schematic diagram of an algorithm for implementing the approach as described above. Fig. 5 shows the processing of the steps and procedures of the method described above, considering a set of traffic requests 5. The traffic requests 5 are split into respective sub-requests 7, each having a determined discrete packet size.

[0075] The preprocessed subrequests 7 are then input to an algorithmic procedure in the quantum conceptual processor 6. For example, the quantum conceptual processor 6 according to Fig. 5 is configured to solve the optimization problem by quantum annealing emulation. The quantum conceptual processor 6 applies the mathematical formulation of the overall optimization problem according to Fig. 4A and Fig. 4B. The quantum conceptual processor 6 then calculates an optimized routing solution of the optimization problem according to Fig. 4B for the subrequests 7, taking into account the constraints according to Fig. 4A.

[0076] After the algorithm procedure is completed, the final calculated minimum value of the optimization problem according to Figure 4B is output from the quantum conceptual processor 6 for each sub-request 7. The communication path k determined by the found optimum value of the optimization problem is then stored for the sub-request 7. The algorithm then ends.

[0077] Therefore, by applying the computer-implemented algorithm procedure according to FIG. 5 based on the above implementation and description regarding FIGS. 2B to 4B, optimized routing can be provided for all traffic demands on individually selected communication paths through the communication network 1.

[0078] The formulation of the optimization problem as a QUBO representation has elegant properties with respect to quantum computing as applied herein within processor 6. Today, quantum computing is still at a significant limit. However, as computer science increasingly advances toward quantum computers, the approach described herein may be further enhanced and developed in the future. For example, as quantum computing becomes increasingly applicable to increasing complexity of the underlying optimization problem, the path box may have more alternative options for communication paths because more path variables can be calculated through quantum computing. Furthermore, as quantum computing becomes increasingly applicable, an increasingly large number of Q-bits, increasingly complex optimization problems, and / or increasingly nonlinear constraints may be considered by the approach described herein.

[0079] The approach described herein is primarily applicable to communication networks, but it can also be applied to any other network, such as a rail network, an energy grid, or a transportation network, where certain "traffic" or "load" must be transmitted across the network via optimized paths.

[0080] The embodiments shown and described herein are examples only. [Explanation of symbols]

[0081] 1. Communication Network 2, 2a~2f communication nodes 3, 3a~3f Aggregation nodes 4 Connections between adjacent nodes 5 Sets of Traffic Requests 5a~5c Traffic requirements 6. Quantum Concept Processor 7 Subrequests d destination node e, e1, e2 edges i1, i2 intermediate nodes o Origin node p subrequests with a particular size p k, k1~k2 Possible communication paths

Claims

1. 1. A computer-implemented method for optimizing utilization distribution in a communication network (1) through which data traffic is routed, the communication network (1) having a plurality of communication nodes connectable via edges (4) of communication paths for routing the data traffic, the method comprising: capturing a set of traffic requests (5), each traffic request (5a-5c) specifying the transfer of a determined data volume from an origin node (o) to a destination node (d) of said plurality of communication nodes (2); splitting said traffic requests (5a-5c) into sub-requests (7,p); specifying a set of alternative communication paths (k) for individual routing of each sub-request (7, p), wherein edges (e) in the set of alternative communication paths (k) are assigned respective utilization capacity limits; for each sub-request (7, p), calculating a fractional capacity utilization of the edges (e) in the set of alternative communication paths (k), the fractional capacity utilization being calculated based on the respective utilization capacity limits; formulating the calculated fractional capacity utilization in terms of a quadratic stress function; and determining an optimized routing by selecting, for each sub-request (7, p), one communication path (k1, k2) from the set of alternative communication paths (k) such that the quadratic stress function is minimized, using a quantum conceptual processor (6).

2. specifying a set of path variables, each path variable being associated with one of the sub-requests (7, p) and one communication path (k1, k2) from said set of alternative communication paths (k); formulating a path term in the quadratic stress function, the path term connecting the calculated fractional capacity utilization of the edge (e) of each communication path (k1, k2) from the set of alternative communication paths (k) with the path variable associated with the each communication path (k1, k2) from the set of alternative communication paths (k); 2. The method of claim 1, further comprising: using the quantum conceptual processor (6) to calculate, for each sub-request (7, p), a path term for selecting one communication path (k1, k2) from the set of alternative communication paths (k) such that the quadratic stress function is minimized.

3. 3. The method of claim 2, wherein the path term is calculated taking into account a path condition that each sub-request (7, p) is routed along exactly one communication path (k1, k2) from the set of alternative communication paths (k).

4. 2. The method of claim 1, wherein a traffic request (5) is split into sub-requests (7, p) having determined discrete data volumes.

5. The quadratic stress function is based on the following constraints on the set of traffic demands (5) or on each of the sub-demands (7, p): the organization of a communication network (1) in different network domains; 2. The method of claim 1, wherein the method is formulated taking into account one or both of the following:

6. The set of alternative communication paths (k) for the individual routing of each sub-request (7, p) is determined by the following constraints: one or more redundant alternative communication paths (k) associated with a sub-network of said communication network (1); the organization of a communication network (1) in different network domains; 2. The method of claim 1, wherein the specified time is determined taking into account one or more of: the latency of the communication network (1);

7. 2. The method of claim 1, wherein the set of alternative communication paths (k) for individual routing of each sub-request (7, p) is specified such that a smaller number of alternative communication paths (k) are selected for topologically close origin and destination nodes (o, d) than for topologically distant origin and destination nodes (o, d).

8. The method of claim 1 , wherein the quadratic stress function is formulated as a quadratic unconstrained binary optimal (QUBO) function.

9. A quantum conceptual processor (6) configured to perform at least the following steps of the method of claim 1: determining an optimized routing by selecting, for each sub-request (7, p), one communication path (k1, k2) from the set of alternative communication paths (k) such that the quadratic stress function is minimized.

10. 10. A computer program comprising instructions, which, when executed by one or more processors, cause the one or more processors to perform the method of claim 1.

11. A computer-readable storage medium having the computer program according to claim 10 stored thereon.

12. A workplace for a network planner configured to validate an optimized routing determined by the method of claim 1.

13. 10. An interface arrangement to a plurality of communication nodes (2) of a communication network (1) through which data traffic is routed, the interface arrangement being configured to automatically deploy an optimized routing determined by the method of claim 1 to the communication nodes (2) of the communication network (1).

Images