Method and system for ensuring time consistency of computer cluster nodes

Abstract

The invention discloses a method and system for ensuring time consistency of computer cluster nodes, and belongs to the field of computer system management. In a cluster system, each node is provided with a time transmit-receive module and a time processing module, system time data is requested from the time processing module of any node to the time transmit-receive modules of all nodes, the time transmit-receive modules of all nodes read the system time data of themselves and send the system time data to the time processing module initiating the request, the time processing module initiating the request broadcasts the processed data to the transmit-receive modules of all nodes after receiving the time information, the transmit-receive modules of all nodes obtain the time data broadcasted by the time processing module and modify the current time of themselves. According to the invention, the time data in the cluster is read and adjusted, the time consistency of the cluster system is ensured without the connection with an external network, and the time of the cluster system is synchronous with the external network.

Description

technical field [0001] The invention discloses a method and system for ensuring time consistency of computer cluster nodes, belonging to the field of computer system management. Background technique [0002] With the development of high-performance computing in China, small, medium, and even large supercomputers are currently deployed in many universities, research institutes, supercomputing centers, and enterprises. These supercomputers range from dozens of computing nodes to tens of thousands of computing nodes. In the process of using the cluster, the time on the node is used as an important information, which is widely used in the running of the program. Some specific applications, such as hadoop, have high requirements on the time synchronization of the cluster. It is very important and necessary to ensure the time consistency of nodes in the cluster. The time of each computing node is managed and recorded by its motherboard chip. When the CMOS battery on the main b...

AI Technical Summary

Problems solved by technology

When the CMOS battery on the main board has problems or failures, the main board time will be disordered, which may easily cause the cluster system to fail

And the system time controlled by the CMOS battery on the current computer motherboard runs for a long time, the system time on most nodes may be slightly faster or slightly slower than the current external real time, and the system time of the node will be much faster due to system failure greater than or significantly slower than the current external real time

However, the current time synchronization method generally uses a single calibration source / calibration node, and the calibration source / calibration node determines the time by accessing a remote network time server. If the time of the source is inaccurate, the time of all machines will be inaccurate; If the cluster is not connected to the external network due to security reasons, the time cannot be calibrated by connecting to a remote time server

Images