A Decentralized POW Computing Power Cluster Deployment Method

Abstract

The invention discloses a decentralized PoW computing power cluster deployment method, including a DRCCP communication process for realizing the deployment of the DRCCP communication protocol. The communication process includes cluster initialization, communication establishment, request synchronization, responsibility statement, fault migration, request forwarding and The cluster is down. Compared with the prior art, the present invention enables each computing node in the cluster to independently coordinate data synchronization, and automatically perform fault migration, effectively reducing the number of nodes, and reducing deployment costs on the premise of realizing fault-tolerant performance; at the same time, through the service The self-scaling of the discovery form can achieve more flexible effects in horizontal expansion.

Description

technical field [0001] The invention relates to the field of distributed networks and consensus mechanisms, in particular to a design scheme for a PoW computing power cluster. Background technique [0002] With the large-scale development of deep learning models, in order to meet the high computing power requirements of deep learning, distributed technology solutions have gradually become a research hotspot. When faced with the huge resource pool of deep learning, it can split the task into multiple subtasks, effectively solve the time-consuming problem in traditional parallel computing, and achieve high-speed and high-efficiency computing effects. In the distributed technology solution, the computing power and processing capability are the core to realize the deployment of the cluster solution. [0003] Distributed technology solutions usually adopt master / slave and multi-master architectures. On the premise of realizing effective cluster scheduling, on the one hand, the ...

AI Technical Summary

Problems solved by technology

When the master node (Master / Leader) goes offline, the slave nodes (Slave) in the cluster will be forced to re-elect a new master node (Master / Leader). At this time, there will be a vacuum period in the cluster, which is not conducive to the development of the cluster work smoothly

As the number of levels increases, the centralized model needs to add additional nodes, which will lead to an increase in the number of master nodes (Master / Leader), which not only wastes the processing performance of some nodes, but also requires additional Leader) to assign

For example, in the election mode, all nodes are actually in a suspended state, which greatly prolongs the waiting time of nodes, wastes the performance of idle nodes, reduces the overall operating efficiency, and makes deployment complex and inefficient

Images