With the advancement of technology, the recent surge in multimedia applications such as video conferencing, Internet telephony, streaming and online gaming may have different sets of QoS requirements, but network domains may also have different policies and modes of implementing Quality of Service (QoS).
In a multi-domain network, providing end-to-end QoS guarantees between all domains where QoS traffic intersects requires strong and satisfactory controls for all members of the network.
However, the dynamic change of the network and the domain manager do not disclose information for reasons of security, etc., and there is a need for a method to provide QoS outside the management area due to the imbalance of device configuration information.
We introduce existing studies considering single domain and multi-domain and introduced QoS support as a challenge in a multi-domain environment. In addition, we looked at cases where reinforcement learning was applied to SDN QoS.
In this dissertation, to provide end-to-end QoS in an SDN environment, stocastic-based end-to-end effective delay is measured, and DAG is applied to determine a flow path, and QoS is applied to SAC-based reinforcement learning as a method to apply QoS in the absence of information. An optimization method was proposed and verified through simulation.
Next, we identified problems when applying reinforcement learning in a multi-domain environment to support QoS in limited information in a multi-domain environment.
We proposed a multi-agent SAC QoS and flow optimization structure to obtain flow optimization while satisfying QoS requirements by using multi-agent SAC that exchanges state between agents to support QoS only with limited information.
Experimental results show that the proposed method can be used as a metric supporting QoS in a multi-domain environment, and can be applied to a QoS-supported learning agent multi-domain environment.
