Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 2

Title:

Deep_Reinforcement_Learning_for_Controller_Placement_i
n_Software_Defined_Network

Summary:
This paper proposes a Deep Reinforcement Learning (DRL) based
approach called Deep Q-Network for Controller Placement (D4CPP) to
optimally place controllers and dynamically adjust the switch-controller
mapping in Software Defined Networks (SDN). The authors model the
Controller Placement Problem (CPP) into a multi-objective joint
optimization problem, where the objectives considered are the latency
between switches and their corresponding controllers, and the load
balance among the distributed controllers. The authors present a detailed
system model and problem formulation along with simulation results to
show the effectiveness of their proposed scheme. The authors claim that
D4CPP can outperform traditional schemes considerably, which confirms
the potential of DRL-based approaches in solving SDN problems.

Strong side:
(1) The authors proposed a practical and efficient approach using Deep
Reinforcement Learning (DRL) for the Controller Placement Problem
(CPP) in Software Defined Networks (SDN).
(2) The authors modeled CPP into a multi-objective optimization
problem, taking into account the latency between switches and their
corresponding controllers and the load balance among distributed
controllers.
(3) The authors evaluated their proposed scheme with extensive
simulations and compared it with traditional schemes. The simulation
results show that the performance of the proposed scheme is better than
the traditional schemes.

Weak side:
(1) The paper does not provide any implementation details.
(2) The simulation results are presented on a small topology, which raises
the question of scalability to a larger and more complex network.
(3) The authors did not discuss the computational overhead incurred by
their proposed scheme.
Alternative way:
An alternative to DRL-based approaches for the CPP problem in SDN
can be the use of meta-heuristic algorithms such as Genetic Algorithms
(GA) or Ant Colony Optimization (ACO). The use of meta-heuristic
algorithms has proved to be effective in solving optimization problems
and can provide a good alternative to DRL-based approaches without
compromising on the quality of the solution. Additionally, meta-heuristic
algorithms have the advantage of being easier to implement and
computationally less demanding as compared to DRL-based approaches.

Terms:
- Software-defined network (SDN)
- Controller placement problem (CPP)
- Deep reinforcement learning
- Multi-controller resource management
- Cooperative content caching
- Vehicular edge computing
- Round-trip time (RTT)
- Single-trip time (STT)
- Queuing latency
- Load balance
- K-means clustering
- Open Science, Scholarship and Services Exchange (OS3E)

Abbreviations:
- D4CPP: Deep Reinforcement Learning for Controller Placement in
Software Defined Network
- DQN: Deep Q-Network
- NP-hard: Non-deterministic Polynomial-time hard
- OS3E: Open Science, Scholarship and Services Exchange
- RTT: Round-Trip Time
- SDN: Software Defined Network
- STT: Single-Trip Time

You might also like