A Performance Prediction Method for Distributed Applications in Cloud Computing Environment

Abstract

The invention discloses a distributed application performance prediction method in a cloud computing environment. Step 1. Perform data training, that is, collect indicators of each container resource layer, micro-architecture layer, and multi-application mixed operation under the separate operation of all applications. The real performance degradation after interference is used as the training data set of machine learning, and the training data set is characterized by indicators, and then the prediction model between the performance index and the application performance is established; step 2, the application is aimed at the performance index and each The "spatial-temporal overlap" encoding information between application components is input into the trained model, so as to obtain the performance degradation prediction results of the application under the interference of mixed parts. Compared with the prior art, the present invention can complete the predictive machine learning algorithm within 4 milliseconds, and achieve an accuracy of 98.48%; it makes up for the shortcomings of related technologies, and can be used for multi-distribution in various scenarios Performance Prediction for Application Disturbances.

Description

technical field [0001] The invention relates to the technical fields of big data and cloud computing, in particular to the task scheduling technology of distributed computing. Background technique [0002] Cloud computing has developed rapidly in recent years. However, with the increase of business on the cloud, the resource utilization rate of the system is not high. In the current technology, placing cloud workloads through mixing is an effective way to improve the utilization of cloud resources. However, due to serious resource contention among workloads, especially for LC (latency-critical) services, mixed loads in servers can easily lead to violations of QoS (Quality of Service). Therefore, it is very important to predict the application performance under the mixed load of cloud computing environment. [0003] Since accurate performance prediction helps to efficiently schedule mixed workloads, there have been extensive discussions on the performance prediction proble...

AI Technical Summary

Problems solved by technology

Almost all work is limited to building interference prediction models on a single server, so existing performance prediction methods are not practical enough

As the complexity and scale of cloud workloads continue to increase, it is very common to deploy workloads in a distributed manner. Existing models are still unable to deal with the interference caused by "inconsistent service component sensitivity" and "temporal and spatial overlap" of distributed workloads. question

[0004] Most of the existing methods cannot be applied to the spatio-temporal interference problem in the most popular distributed application scenarios. Most of the existing performance prediction models are aimed at the "full overlapping" interference scenarios, and they ignore the "partial overlapping" interference scenarios, especially For distributed cloud services, the "partial overlap" interference scenario is particularly evident

Images