Non-invasive server energy consumption decomposition system and method

Abstract

The present invention discloses a non-invasive server power consumption decomposition system. The system specifically comprises: obtaining overall energy consumption of all servers and a state of each server in a cabinet, and by using a machine learning manner, obtaining a correspondence between the server state and energy consumption, so as to decompose energy consumption of each server from the overall energy consumption. The system has the characteristics of a low cost, high accuracy, easy deployment, and the like. The system comprises four modules: a state collection module, taking charge of running a monitor application on each server in the cabinet so as to collect server states; an energy consumption monitoring module, taking charge of reading overall energy consumption of the cabinet from the PDU; a third module, taking charge of taking information obtained by the first two modules as a data set training model so as to decompose energy consumption of each server; and the fourth module, taking charge of continually increasing the comprehensiveness of the data set so as to make the decomposed result more accurate.

Description

technical field [0001] The invention belongs to the field of green computing, and more specifically relates to a non-intrusive system and method for decomposing energy consumption of servers. Background technique [0002] With the development of high-performance computing, large-scale server clusters are becoming more and more common. Due to space and cost constraints, more and more clusters use blade servers. This blade server has the advantages of small size, high computing density, low power consumption, low cost, and easy deployment and expansion. But at the same time, due to the pursuit of high computing density, many components of the blade server are also highly integrated, such as switching modules, power supply modules, and heat dissipation modules, etc. are all integrated in the blade box. The power supply of the blade server in the knife box comes from multiple sets of mutually redundant power supply modules, that is, multiple power supply modules supply power t...

AI Technical Summary

Problems solved by technology

[0003] Aiming at the above defects or improvement needs of the prior art, the present invention provides a non-intrusive blade server energy consumption decomposition system and method, which monitors the power consumption of the entire cabinet and combines the resource utilization of each blade server to utilize machine The learning method obtains the corresponding relationship between server status and power consumption, so as to accurately decompose the power consumption of each blade server, and then solve the large cost increase caused by configuring hardware devices for each server separately in the existing method technical issues

Images