Modeling Server Workloads for Campus Email Traffic Using Recurrent Neural Networks

Abstract

As email workloads keep rising, email servers need to handle this explosive growth while offering good quality of service to users. In this work, we focus on modeling the workload of the email servers of four universities (2 from Greece, 1 from the UK, 1 from Australia). We model all types of email traffic, including user and system emails, as well as spam. We initially tested some of the most popular distributions for workload characterization and used statistical tests to evaluate our findings. The significant differences in the prediction accuracy results for the four datasets led us to investigate the use of a Recurrent Neural Network (RNN) as time series modeling to model the server workload, which is a first for such a problem. Our results show that the use of RNN modeling leads in most cases to high modeling accuracy for all four campus email traffic datasets.

Modelling email traffic workloads with RNN and LSTM models

“…In order to address the limitations of probabilistic models, email traffic was evaluated as a time series problem using Recurrent Neural Networks in [5] and the prediction accuracy was found to be substantially higher than the probabilistic modelling approach in [6]. In this work we provide further investigation with the use of tuned hyper-parameters.…”

“…RNN and LSTM models are evaluated to propose a model which could best fit all email traffic categories. Boukoros, et al [5] Focus on modelling email traffic as a time series problem. The datasets were collected from four universities over several months…”

“…The authors did not investigate the time series properties of email traffic data. The work in [5] used a RNN model to model email traffic workloads as a time series, including the outliers [5]. However, the authors did not evaluate the best fit hyper-parameters for the categories of email traffic to overcome the vanishing gradient problem.…”

“…The sampled time series data often contain noise and have high dimensionality properties, which make them challenging to analyse and model with conventional probabilistic techniques. In order to overcome the limitations of probabilistic models, email traffic was modelled using RNNs in [5], treating email traffic workload as a time series problem. The prediction accuracy was found to be substantially higher than that of the probabilistic modelling approach in [6].…”

“…The main objectives of this work are to: (a) present and discuss the relevant literature on email traffic modelling, (b) develop a comprehensive structured approach to maximize the generalization ability of a RNN model to avoid the vanishing gradient problem when using backpropagation. The LSTM model [7] and a different implementation of the RNN model compared to the one used in [5] are investigated in our work to examine if they can provide efficient modelling for all types of emails in our datasets.…”