Data communication through Edge devices in a secured channel for real time applications is one of the biggest concerns. Software Defined Network is a suitable network opted for security. The Control Plane is more vulnerable to variety of attacks. Ensemble machine learning approach which composes of 3 times of Random Forest with one time of Linear Regression gives us the prediction of errors and loss of packets and hence the data will be transferred more securely. The Block Chain is integrated with the SDN controllers in the control plane to transfer the data from the edge devices in the form of a Block to the cloud layer. Federated Learning can be inculcated for security analysis prediction and then aggregating into a Global Model in the Central Cloud Layer. The majority of contemporary FL techniques do not explicitly address variances among client parameter estimates. An aggregation mechanism built on the Hessian matrix in order to close the gap. Moreover, we can use the Hessian as a scaling matrix using the second -order partial derivative information of the loss function
Top1. Introduction
A Software Defined Network (SDN) is a virtual, programmable networking architecture which is primarily meant for detecting the behaviour of the entire network. SDN is a dynamic architecture and hence it has its unique attention towards industry and research environment. Due to its scalability, adaptability and less structural complexity, the SDN has its wide usage in various real time applications recently. Apart from these, it has a higher privilege of building a securable environment for communication of data packets when compared to traditional TCP/IT network. SDN is broadly classified into three planes: Application plane, Control plane and Data plane. Though SDN overcomes several security limitations of traditional networks, it still faces some security breaches and has a vulnerable projection in Control plane. Due to security threats in the Control plane, there are still few packet losses. As a result, some data are left unnoticeable, could not be taken into account and it has a serious effect on final project conclusions.
This chapter concentrates on overcoming the security concerns in the Control plane and the techniques proposed to eradicate the unseen challenges. SDN builds a novel security mechanism by encompassing Blockchain based data packet transmission. The presence of SDN Controller in the Control plane manages the network traffic and aims at regulating the packets based on the application request from the Application plane.
Blockchain building a secured architecture by integrating with SDN, can be able to strike a good balance between scalability, irrespective of the size of the query from the Application plane and at the performance level, throughput and latency metrics are perfectly achieved. Every node in the network separately assembles the BC. Each block includes information on some or all of the most recent transactions that were not included in the block before it. Each block is made up of two parts: (i) a block header with the current and previous cryptographic block hash values, a timestamp, and a nonce; and (ii) the main body with the transaction hash value, sender and receiver identities, and signatures for each transaction. Block chain is the primary aspect of providing security against threats, where it could be accomplished by Miners. Miners are users who pool their computing resources to check the validity of the transactions contained in blocks in exchange for rewards. To confirm the most recent transactions and record them all in the global ledger, they must solve a statistical problem based on the computation of a cryptographic hash function. A block is typically extracted every five to ten minutes.
Federated Learning (FL) approach is the most preferable ensemble based prediction learning model since the training data remains distributed locally (Nguyen et al., 2021) and demands strict privacy over the data. Federated learning (FL) is a distributed AI strategy that works by coordinating several devices to do AI training without sharing raw data in order to protect privacy and conserve network resources (such bandwidth).This would prominently reduce the probability of data transmission for longer distances which in turn security threats are minimized. The local training model collects the data from the nearby Edge devices; convert into a block, thus supporting the decentralization, immutability and traceability. Decentralised data ledgers can be used to achieve federated learning without the need for a central server, which reduces the risk of single-point failures. The unexpected update of events and dynamic network behaviours are all being tracked by the network entities. This would probably mitigate the Distributed Denial of Attacks (DDoA). The integration of FL and Blockchain enhances trust among the devices and improves the scalability of intelligent Edge networks.
FL is an iterative improvement model, where the local model is trained over the data in a continuous iteration process until the model is completely trained and the prediction performance is highly achieved. At each and every iteration, all the local updates are sent back to the cloud and the aggregation followed by final prediction is accomplished in Cloud.