GPU monitoring alarm system with monitoring function customized by cloud platform

Abstract

The invention discloses a GPU monitoring alarm system with a monitoring function customized by a cloud platform, and belongs to the monitoring alarm technical field of the cloud platforms. The technical problem to be solved by the invention is how to realize the user-defined monitoring configuration according to the requirements, and generate the monitoring data satisfying the requirements of theusers flexibly. According to the technical scheme, the system comprises a data acquisition module, a monitoring configuration management module, an alarm rule management module and a data processingmodule; the data acquisition module is used for periodically acquiring the performance indexes of a GPU; the monitoring configuration management module is used for configuring a GPU monitoring dimension, a GPU monitoring index, a GPU monitoring period and a GPU monitoring statistical method; the alarm rule management module is used for configuring the alarm rules; the data processing module is used for storing the acquired data and generating the monitoring data according to the monitoring configuration and the acquired data and is used for traversing the alarm rules regularly, generating thealarm data or clearing the alarm data according to the acquired data, and forwarding the alarm data according to a configured notification mode.

Description

technical field [0001] The invention relates to the technical field of monitoring and alarming of cloud platforms, in particular to a GPU monitoring and alarming system capable of custom monitoring of cloud platforms. Background technique [0002] For three decades, changes in CPU performance have never deviated from Moore's Law. But improvements in CPU performance have slowed. GPU computing defines a new overload law. It starts with highly specialized parallel processors and continues to evolve through system design, system software, algorithms, and optimized applications. It is especially suitable for the increasing demand for computing power in application scenarios such as artificial intelligence, HPC, and graphics and image processing. [0003] GPU cloud physical hosts in the form of bare metal can provide computing power of "one machine with multiple cards" or "multiple machines with multiple cards". However, for some users, multiple GPU cards exceed the user's com...

AI Technical Summary

Problems solved by technology

This technical solution can observe GPU performance in real time, which is convenient for managers to grasp the usage of GPU resources. In the case of GPU overload, they can receive notification information in time, allocate resources, and make reasonable responses, which greatly reduces the management of the entire workstation. Maintenance costs improve the efficiency of administrator maintenance, but it is not possible for users to customize the monitoring configuration according to their needs and flexibly generate monitoring data that meets user needs

Images