Research On Message Distribution Algorithm For Peer-to-Peer Distributed Systems

With the development of P2P technology,peer-to-peer distributed systems are gradually replacing traditional distributed systems.Due to its decentralized nature,messages are no longer distributed through a central server,so it is a great challenge to design a decentralized message distribution algorithm.In the traditional centralized "client/server" architecture,messages can be easily distributed only through the central server,while for decentralized systems,a specific message distribution protocol is required.In the process of message distribution,important performance indicators such as distribution rate,bandwidth occupancy,reliability,security,and load balancing need to be considered.In different distributed systems,there must be targeted message distribution protocols according to different needs.Gossip algorithm is a classic message distribution algorithm in point-to-point architecture.Its advantages of simple fault tolerance and scalability are very popular in many systems,but its shortcomings of high redundancy and high overhead still need attention.In view of the above problems,this paper conducts targeted research and improvement.The innovations of this paper are as follows:(1)The NHD-Gossip message distribution algorithm is proposed to improve the distribution speed of the Gossip algorithm based on rumor propagation,and change the original fixed gossip factor to a dynamic decaying gossip factor.Three groups of networks of different scales are used for experimental analysis,and the results show that under the condition of ensuring the same overall convergence speed as the original algorithm,the speed of receiving messages can be accelerated by more than 64%of nodes,and the algorithm overhead and execution time are compared through simulation experiments.(2)The HRM-Gossip transmission topology self-optimization algorithm is proposed,aiming at the problem of unbalanced message distribution load caused by the random structure of the unstructured P2P overlay network,and using the Gossip algorithm distribution process to continuously optimize the transmission topology.A P2P node coldness score function is proposed,and topology adjustment is performed according to the coldness ranking,and finally the coldness and heatness of the whole network nodes are evenly distributed to achieve the optimization purpose of message distribution load balancing.Experiments show that using the Gossip algorithm to distribute messages on the optimized overlay network speeds up the convergence speed,reduces network overhead,and balances node load compared to before optimization.In terms of topology robustness,the experiments of frequent nodes joining and exiting the network show that the optimized topology still maintains strong connectivity,reflecting the high fault tolerance and scalability of the topology.(3)Combining the first two algorithms,a complete message distribution framework under a large-scale point-to-point distributed system is formed,and a comparative experiment is carried out in combination with other member management protocols.The results show that the convergence speed of the message distribution algorithm in this paper is increased by up to 41.67%.In the case of frequent network jitter,it can quickly restore the load balance state,reflecting the advantages of robustness,fault tolerance and scalability.