Method for ensuring strong consistency of data in distributed system

Abstract

According to a method, a reliable multicast mechanism is adopted, data is distributed from a source business cluster to a target business cluster, at any time, the target cluster only has one node (named as a main node) which is responsible for distributing processed business data to the outside again, and other nodes are named as standby nodes. All the nodes in the target business cluster join the same multicast group and process the received service data at the same time, but the standby node does not participate in external data distribution, and the processing speeds of all the nodes are consistent, so that the conversion sequence of internal state machines can be kept consistent, and the strong data consistency among the nodes is ensured. If the main node fails, the reasonable standby node can be selected to be promoted to the main node, and data processing is continued, so that distributed data consistency is achieved, namely, all data can be accessed from any node in the cluster. According to the invention, the original TCP point-to-point transmission mode is replaced by connectionless reliable multicast, so that the number of nodes accommodated by a service cluster is obviously increased.

Description

technical field [0001] The invention belongs to the technical field of securities trading, and in particular relates to a method for ensuring strong consistency of data in a distributed system. Background technique [0002] It has been successfully applied in multiple product lines such as securities, futures, and institutional trading. However, with the rapid growth of trading varieties and trading volume in the financial market, higher requirements are put forward for the performance and capacity of the trading system. However, there are many limitations in the original trading system, which makes it very difficult to improve and improve on the basis of the existing version, and then puts forward the requirements for a new generation of trading system. [0003] For example, the original transaction system is not a fully distributed architecture. The business cluster only supports a simple master-standby model, and the master-standby switching mechanism itself is controlle...

AI Technical Summary

Problems solved by technology

When the arbitration node fails, it means that the switching function of the active and standby nodes is unavailable. When the active node fails and cannot provide transaction services, the cluster actually fails, and the business stability and reliability are not good.

[0006] (2) The business cluster has only two active and standby nodes, and the horizontal expansion capability is limited. If the customer requires to increase transaction nodes, maintain data consistency, and perform disaster recovery switching at any time, it is obvious that this architecture cannot meet the customer's further demand for fault tolerance.

Images