Server surface heat condition-based operation state diagnosis method

Abstract

The invention provides a server surface heat condition-based operation state diagnosis method. The method comprises the following steps of: acquiring an infrared image of a server under an operation state through a thermal infrared imager; carrying out standard tilt correction processing to obtain a server interested area; obtaining a server hotspot map, namely, an area, with relatively high relative temperature, of the server, by adoption of a homomorphic filtering enhanced image and a one-dimensional maximum entropy; extracting morphological features and texture features of the hotspot map;and finally inputting obtained feature data into a support vector machine classifier so as to obtain a server classification result. According to the method, the disadvantages of contact type server diagnosis methods are effectively solved, and the efficiency and correctness of checking server hotspot areas, namely, local heating areas, by data center managers are effectively enhanced. The methodis suitable for monitoring server operation states.

Description

technical field [0001] The invention belongs to the technical field of signal and information processing, and in particular relates to a method for diagnosing the running state based on the thermal state of the server surface. Background technique [0002] In recent years, with the rise of cloud computing and its applications, the cloud computing data center (Cloud computing Data Center, CDC) has also been vigorously developed. In 2015, the Ministry of Industry and Information Technology, together with the National Energy Administration and the State Organ Affairs Administration, issued the "National Green Data Center Pilot Work Plan" (hereinafter referred to as the plan). Solution disclosure: my country's data centers are developing rapidly, with a total of more than 400,000, and the annual power consumption exceeds 1.5% of the electricity consumption of the whole society. The average Power Usage Effectiveness (PUE) of most data centers is About 1.89, there is a big gap com...

AI Technical Summary

Problems solved by technology

Early measurement methods used wired sensors, but expensive installation and configuration costs prevented widespread use

Although wireless sensors have the advantages of low cost and non-invasive measurement, Chieh-Jan MikeLiang et al pointed out in the paper "RACNet: A High-Fidelity Data Center Sensing Network" that the electromagnetic environment of the data center is not conducive to the large-scale wireless sensor network. communication

Images