No abstract

No abstract

Science networks and their hosted applications require large and frequent data transfers, but these transfers are subject to network performance degradation, including queuing delays and packet drops. However, well known network dynamics along with limited instrumentation access complicate the creation of an accurate method that predicts different performance aspects of data transfers. In this study, we develop a lightweight machine learning tool to predict end-to-end packet retransmission in science flows of arbitrary size. We also identify the minimum set of necessary path and host measurements needed as input features in our predictor in order to achieve high accuracy. In our evaluation process our predictor demonstrated low training times and was able to provide accurate estimates (97-99%) for packet retransmissions of data transfers of arbitrary sizes. The results also manifest that the our solution was able to predict retransmit behavior reasonably well (66%) even for previously unseen data if training and testing datasets had similar statistics.

Abstract-Traditional intrusion detection systems are not adaptive enough to cope with the dynamic characteristics of cloud-hosted virtual infrastructures. This makes them unable to address new cloud-oriented security issues. In this paper we introduce SAIDS, a self-adaptable intrusion detection system tailored for cloud environments. SAIDS is designed to reconfigure its components based on environmental changes. A prototype of SAIDS is described.