Prediction based Dynamic Load Balancing System for Text to Speech Conversion

An Implementation for a Prediction based Internet Load balancing Algorithm

“…Most of our work in this paper builds on top the work done by Patil et al [1]. However, there is immense research related to the area of Predictive load balancing and some of the related papers are highlighted in this section.…”

“…Patil et al [1] designed a predictive load balancing algorithm for the use case of a Text-to-Speech synthesis application. Their system follows a Master-Slave architecture where the load balancer breaks text input from each request into individual lines and each line is sent to the worker node with minimum current load.…”

“…Predictive load balancing algorithms aim to remove this requirement by trying to predict load induced on servers due to requests rather than measuring it directly. We aim to provide an improved implementation of algorithm described by Patil et al [1] and compare this implementation with a static algorithm like Round Robin in terms of performance and resource utilisation. This implementation is for a web application which does text-to-speech synthesis.…”

An Implementation for a Prediction based Internet Load balancing Algorithm

“…Our proposed system follows the same Master-Slave design as described by Patil et al [1]. However, we have developed a general equation for predicting synthesis time for each server by taking into account information like number of CPU cores, RAM and GPU availability of each server rather than specific linear regression models trained for each server to the pool as described in [1]. We have also compared performance of various predictive models for predicting synthesis time.…”

“…The actual request distribution process is described by Algorithm 1. We chose the Lasso Regression model for predicting synthesis time as it provided a lower error rate on our dataset than LinearRegression(chosen by the authors of [1]). The dataset was generated manually using the PredictionLogs table described earlier.…”