Instance expansion algorithm for micro-service with prediction
Instance expansion of micro-service consumes time for constructing new instances, it cannot satisfy the requirement of low latency service, such as scientific calculation workflows. In order to reduce the time for expanding instances, an instance expansion algorithm for micro-service with prediction is proposed in this letter, which sets correntropy as cost function and uses maximum correntropy criteria to filter the burst service requirements to improve the accuracy of prediction, and use stochastic gradient descent algorithm to train the data set to predict the required micro-service instances in the next time. The performance of the proposed algorithm is analysed in a real experiment telecom office with the compared ones using A/B test, and the experimental results show that the proposed algorithm has 70% less time of instance expansion than the compared ones, and the accuracy of the proposed algorithm is 20% more than the compared one which uses least mean square as the cost function.
