System and method to learn and prescribe network path for SDN
A path suggestion tool in a Software-Defined Networking (SDN) architecture to predict a router's future usage based on an analysis of the router's historical usage over a given period of time in the past and to recommend a routing path within the network in view of the predicted future usages of the routers/switches in the network. The path suggestion tool is an analytical, plug-and-play model usable as part of an SDN controller to provide more insights into different routing paths based on the future usage of each router. A Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) model in the suggestion tool analyzes the historical usage data of a router to predict its future usage. A Deep Boltzmann Machine (DBM) model in the suggestion tool recommends a routing path within the SDN-based network upon analysis of the LSTM-RNN based predicted future usages of routers/switches in the network.
1. A method comprising:
selecting, by a computing device, a Long Short Term Memory Recurrent Neural Network (LSTM-RNN) model, wherein
the LSTM-RNN model includes a Deep Boltzmann Machine (DBM) model having a single visible layer and multiple hidden layers;
modeling, by the computing device, each individual routing element of a plurality of routing elements in a data plane of a Software-Defined Networking (SDN) architecture as a distinct unit in a corresponding one of the hidden layers of the DBM model;
obtaining a historical usage pattern of the individual routing elements of the plurality of routing elements;
training, by the computing device, the DBM model based on the historical usage pattern of individual routing elements of the plurality of routing elements, wherein the training is performed in a control plane of the Software-Defined Networking (SDN) architecture,
wherein the data plane and the control plane of the SDN architecture are decoupled;
providing, by the computing device, a source Internet Protocol (IP) address and a destination IP address of a data packet to be routed through the data plane as units of the visible layer of the DBM model; and
using, by the computing device, the trained DBM model to evaluate predicted future usages of corresponding routing elements forming the hidden layers of the DBM model to recommend a routing path for the data packet within the data plane based on the source and the destination IP addresses.
2. The method of claim 1 , wherein the DBM model is a stacked LSTM-RNN model.
3. The method of claim 1 , further comprising:
predicting, by the computing device, network utilization of the data plane over a pre-defined time period based on a corresponding predicted future usage of each individual routing element of the plurality of routing elements in the data plane.
4. The method of claim 1 , further comprising:
recommending, by the computing device, a routing path within the data plane that is based on a corresponding predicted future usage of each routing element of the plurality of routing elements within the routing path as determined by the DBM model.
5. The method of claim 1 , further comprising:
recommending, by the computing device, a routing path within the data plane based on predicted future usages of all of the plurality of routing elements in the data plane.
6. The method of claim 1 , wherein the computing device is an SDN controller.
7. A computing system comprising:
a memory storing program instructions; and
a processing unit coupled to the memory and operable to execute the program instructions, which, when executed by the processing unit, cause the computing system to:
select a Long Short Term Memory Recurrent Neural Network (LSTM-RNN) model, wherein
the LSTM-RNN model includes a Deep Boltzmann Machine (DBM) model having a single visible layer and multiple hidden layers;
model each individual routing element of a plurality of routing elements in a data plane of a Software-Defined Networking (SDN) architecture as a distinct unit in a corresponding one of the hidden layers of the DBM model;
obtain a historical usage pattern of the individual routing elements of the plurality of routing elements,
wherein the data plane and a control plane of the SDN architecture are decoupled;
train the DBM model based on a historical usage pattern of the individual routing elements of the plurality of routing elements, wherein the training is performed in a control plane of the Software-Defined Networking (SDN) architecture;
provide a source Internet Protocol (IP) address and a destination IP address of a data packet to be routed through the data plane as units of the visible layer of the DBM model; and
use the trained DBM model to evaluate predicted future usages of corresponding routing elements forming the hidden lavers of the DBM model to recommend a routing path for the data packet within the data plane based on the source and the destination IP addresses.
8. The computing system of claim 7 , wherein the DBM model is a stacked LSTM-RNN model.
9. The computing system of claim 7 , wherein the program instructions, upon execution by the processing unit, cause the computing system to:
use the (DBM) DBM model to recommend a routing path for a data packet within the data plane that is based, at least in part, on a corresponding predicted future usage of each individual routing element of the plurality of routing elements within the routing path and based at least in part, on a source IP address and a destination IP address for the data packet.
10. The computing system of claim 7 , wherein the program instructions, upon execution by the processing unit, cause the computing system to:
predict network utilization of the data plane over a pre-defined time period based on a corresponding predicted future usage of each individual routing element of the plurality of routing elements in the data plane.
11. The computing system of claim 7 , wherein the computing system is an SDN controller.
12. A computer program product comprising a non-transitory computer-usable medium having computer-readable program code embodied therein, the computer-readable program code adapted to be executed by a computing system to implement a method comprising:
selecting a Long Short Term Memory Recurrent Neural Network (LSTM-RNN) model, wherein
the LSTM-RNN model includes a Deep Boltzmann Machine (DBM) model having a single visible layer and multiple hidden layers;
modeling each individual routing element of a plurality of routing elements in a data plane of a Software-Defined Networking (SDN) architecture as a distinct unit in a corresponding one of the hidden layers of the DBM model;
obtaining a historical usage pattern of the individual routing elements of the plurality of routing elements,
wherein the data plane and a control plane of the SDN architecture are decoupled;
training the DBM model based on a historical usage pattern of individual routing elements of the plurality of routing elements, wherein the training is performed in a control plane of the Software-Defined Networking (SDN) architecture;
providing a source Internet Protocol (IP) address and a destination IP address of a data packet to be routed through the data plane as units of the visible layer of the DBM model; and
using the trained DBM model to evaluate predicted future usages of corresponding routing elements forming the hidden layers of the DBM model to recommend a routing path for the data packet within the data plane based on the source and the destination IP addresses.
13. The computer program product of claim 12 , wherein the method further comprises:
recommending a routing path within the data plane that is based on a corresponding predicted future usage of each individual routing element of the plurality of individual routing elements within the routing path as determined by the DBM model.
14. The computer program product of claim 13 , wherein the recommending comprises:
using the DBM model to recommend the routing path of a data packet in the data plane based, at least in part, on the predicted future usage of individual routing elements of the plurality of routing elements and, at least in part, on a source IP address and destination IP address for the data packet.
15. The computer program product of claim 12 , wherein the method further comprises:
recommending a routing path within the data plane based on predicted future usages of all individual routing elements of the plurality of routing elements in the data plane.