Mass computing node resource monitoring and management method for high-performance computer

Abstract

The invention discloses a mass computing node resource monitoring management method for a high-performance computer, which comprises the following steps that: enabling a control node to send a messagesending request through an intermediate node: enabling the control node to take out the message sending request and generate a working thread for processing the message sending request; selecting a normal intermediate node through the working thread; forwarding the message sending request to the selected intermediate node through the working thread, then waiting for a message returned by the intermediate node, and skipping to execute the next step after receiving the message returned by the intermediate node; and enabling the working thread to process the returned message, updating the statesof the intermediate node and the computing node, and ending the working thread. According to the invention, a layer of intermediate node is added between the control node and the massive computing nodes to share the load of the control node in the process of monitoring and managing massive computing node resources, and the related load of the computing nodes in the process is reduced at the sametime.

Description

technical field [0001] The invention relates to a high-performance computer massive computing node resource management technology, in particular to a high-performance computer-oriented massive computing node resource monitoring and management method. Background technique [0002] Currently, a management mode in which a single control node controls a large number of computing nodes is adopted for massive computing node resources in high-performance computers. During the operation of the system, the control node needs to monitor and record the real-time status of each computing node for task assignment and other work. The main way to realize this function is that the control node continuously generates a request to send messages to the computing node (message sending request), obtains the current status of the computing node according to the return message of the computing node and modifies the data structure used to manage the computing node on the control node. The common f...

AI Technical Summary

Problems solved by technology

However, this will cause the system to be in a contradictory state: First, if the two upper limits are not changed, when the node size increases, although it can be guaranteed that the number of threads will not exceed the upper limit at the same time, it will make a large number of The performance of the system will be seriously damaged if the sending request enters the waiting state and cannot be processed in time

Secondly, once these two upper limits are increased, although the sending request can be guaranteed to be processed in time, the load on the control node will also increase accordingly

[0009] 2. When the size of the node increases, the load related to message forwarding on the computing node increases

[0011] To sum up, under the condition of increasing node scale, if a tree structure is used to send messages, the width of the communication tree cannot be greatly increased, otherwise it will bring greater load to both the control node and the computing node, reducing the system performance

Images