User plane selection using reinforcement learning
A method of reinforcement learning is used for placement of a plurality of service functions at nodes of a telecommunications network. The state of the system is defined by an allocation matrix, wherein each first vector of the allocation matrix corresponds to a respective one of the nodes of the telecommunications network, each second vector of the allocation matrix corresponds to a respective one of the plurality of service functions. Moreover, each cell of the allocation matrix contains a value 1 if the one of the plurality of service functions corresponding to the respective second vector is placed on the one of the nodes of the telecommunications network corresponding to the respective first vector, and otherwise contains a value 0.
1 . A method of reinforcement learning for placement of a plurality of service functions at nodes of a telecommunications network, the method comprising:
defining a state of a system by means of an allocation matrix, wherein the state of the system includes an indication of service functions supported in the system, wherein:
each first vector of the allocation matrix corresponds to a respective one of the nodes of the telecommunications network,
each second vector of the allocation matrix corresponds to a respective one of the plurality of service functions, and
each cell of the allocation matrix contains a value 1 if the one of the plurality of service functions corresponding to the respective second vector is placed on the one of the nodes of the telecommunications network corresponding to the respective first vector, and otherwise contains a value 0; and
utilizing the allocation matrix as input to train a reinforcement learning (RL) agent to function as a placement algorithm for placing the plurality of service functions among the nodes of the telecommunication network.
2 . The method according to claim 1 , comprising further defining the state of the system by at least one second matrix, wherein one or more of at least one second matrix contains information specific to a family of service sets that share requirements.
3 . The method according to claim 2 , comprising further defining the state of the system by a service function type matrix, wherein:
each first vector of the service function type matrix corresponds to a respective type of service function,
each second vector of the service function type matrix corresponds to a respective one of the plurality of service functions, and
each cell of the service function type matrix contains a value 1 if the one of the plurality of service functions corresponding to the respective second vector comprises a service function of the type corresponding to the respective first vector, and otherwise contains a value 0.
4 . The method according to claim 2 , comprising further defining the state of the system by a key performance indicator matrix, wherein:
each first vector of the key performance indicator matrix corresponds to a respective type of key performance indicator,
each second vector of the key performance indicator matrix corresponds to a respective one of the nodes of the telecommunications network, and
each cell of the key performance indicator matrix contains a value indicating a value of the corresponding key performance indicator for the corresponding one of the nodes of the telecommunications network.
5 . The method according to claim 2 , comprising further defining the state of the system by an ordering matrix, wherein:
each first vector of the ordering matrix corresponds to a respective one of the plurality of service functions,
each second vector of the ordering matrix also corresponds to a respective one of the plurality of service functions, and
each cell of the ordering matrix contains a value 1 if the one of the plurality of service functions corresponding to the respective first vector should be traversed by data passing through the plurality of service functions before the one of the plurality of service functions corresponding to the respective second vector, and otherwise contains a value 0.
6 . The method according to claim 2 , comprising further defining the state of the system by a latency constraint matrix, wherein:
each first vector of the latency goal matrix corresponds to a respective one of the plurality of service functions,
each second vector of the latency goal matrix also corresponds to a respective latency value, and
each cell of the latency goal matrix contains a value 1 if the one of the plurality of service functions corresponding to the respective first vector has a latency requirement corresponding to the latency value of the respective second vector, and otherwise contains a value 0.
7 . The method according to claim 2 , comprising further defining the state of the system by at least one goal matrix, wherein the at least one goal matrix contains information specific to a subset of a family of service sets.
8 . The method according to claim 7 , comprising further defining the state of the system by a latency goal matrix, wherein:
each first vector of the latency goal matrix corresponds to a respective one of the plurality of service functions,
each second vector of the latency goal matrix also corresponds to a respective latency value, and
each cell of the latency goal matrix contains a value 1 if the one of the plurality of service functions corresponding to the respective first vector has a latency requirement corresponding to the latency value of the respective second vector, and otherwise contains a value 0.
9 . The method according to claim 7 , comprising further defining the state of the system by a co-location goal matrix, wherein:
each first vector of the co-location goal matrix corresponds to a respective one of the plurality of service functions,
each second vector of the co-location goal matrix also corresponds to a respective one of the plurality of service functions, and
each cell of the co-location goal matrix contains a value 1 if the one of the plurality of service functions corresponding to the respective row should be co-located with the one of the plurality of service functions corresponding to the respective second vector, and otherwise contains a value 0.
10 . A method of reinforcement learning for placement of a plurality of service functions at nodes of a telecommunications network, the method comprising:
determining a plurality of possible goal matrices;
running a reward calculator for each of the possible goal matrices, to calculate a respective reward value for each of the possible goal matrices;
selecting one of the calculated reward values; and
outputting the selected one of the calculated reward values and the corresponding one of the possible goal matrices as a virtual reward and a virtual goal matrix, wherein the virtual reward and virtual goal matrix are used as input to train a reinforcement learning (RL) agent to function as a placement algorithm for placing the plurality of service functions among the nodes of the telecommunication network.
11 . The method according to claim 10 , wherein the step of selecting one of the calculated reward values comprises selecting a largest reward value of the calculated reward values.
12 . The method according to claim 10 , comprising performing the steps of running the reward calculator for each of the possible goal matrices, selecting one of the calculated reward values, and outputting the selected one of the calculated reward values as a virtual reward and a virtual goal matrix only in response to determining that a cost of running the reward calculator is below a threshold and/or that a dimension of each goal matrix is below a threshold.
13 . A computer program comprising instructions which, when executed on at least one processor, cause the at least one processor to carry out a method according to claim 1 .
14 . A carrier containing a computer program according to claim 13 , wherein the carrier comprises one of an electronic signal, optical signal, radio signal or computer readable storage medium.
15 . A computer program product comprising non transitory computer readable media having stored thereon a computer program according to claim 13 .
16 . Apparatus for performing a method of reinforcement learning for placement of a plurality of service functions at nodes of a telecommunications network, the apparatus comprising a processor and a memory, the memory containing instructions executable by the processor such that the apparatus is operable to:
define a state of a system by an allocation matrix, wherein the state of the system includes an indication of service functions supported in the system, wherein:
each first vector of the allocation matrix corresponds to a respective one of the nodes of the telecommunications network,
each second vector of the allocation matrix corresponds to a respective one of the plurality of service functions, and
each cell of the allocation matrix contains a value 1 if the one of the plurality of service functions corresponding to the respective second vector is placed on the one of the nodes of the telecommunications network corresponding to the respective first vector, and otherwise contains a value 0; and
utilizing the allocation matrix as input to train a reinforcement learning (RL) agent to function as a placement algorithm for placing the plurality of service functions among the nodes of the telecommunication network.
17 . Apparatus for performing a method of reinforcement learning for placement of a plurality of service functions at nodes of a telecommunications network, the apparatus comprising a processor and a memory, the memory containing instructions executable by the processor such that the apparatus is operable to:
determine a plurality of possible goal matrices;
run a reward calculator for each of the possible goal matrices, to calculate a respective reward value for each of the possible goal matrices;
select one of the calculated reward values; and
output the selected one of the calculated reward values and the corresponding one of the possible goal matrices as a virtual reward and a virtual goal matrix, wherein the virtual reward and virtual goal matrix are used as input to train a reinforcement learning (RL) agent to function as a placement algorithm for placing the plurality of service functions among the nodes of the telecommunication network.