Measurement Correlation for Improving Cooperation in Measurement Federations J´ eferson Campos Nobre * Leandro Lisboa Penz † , Lisandro Zambenedetti Granville † , * University of Vale do Rio dos Sinos, S˜ ao Leopoldo – Brazil † Federal University of Rio Grande do Sul, Porto Alegre – Brazil Email: jcnobre@unisinos.br, lpenz@lpenz.org, granville@inf.ufrgs.br Abstract—The diversity of services that operate in the Internet has increased significantly in the last years. Performance prob- lems in these services cause important financial losses. In order to ensure that these problems do not occur, service levels need to be monitored. One of the main techniques for this monitor- ing involves the utilization of active measurement mechanisms. However, such mechanisms are expensive in terms of resources consumption due to the activation of measurement sessions. Thus, such sessions usually can cover only a fraction of what could be measured, which can lead to service level problems being missed. Measurement federations can help network administrators in different tasks, such as the control of the activation of active measurement sessions. In this context, measurement correlation can be deployed in order to improve this control in such federa- tions. The main contribution of the present work is the proposal of a data transformation service which provides measurement correlation. This service can be used to enable cooperation features in measurement federations, while decreasing resource consumption. Besides that, statistical tests that can be used to compose such correlation are presented. The proposed solution provides valuable insights regarding measurement correlation from federated measurement points and can be used for the design of better application for the control of active measurement sessions. I. I NTRODUCTION The capacity and accessibility of networking infrastructures has increased in the last years. Besides that, the diversity of applications and service which operate in such infrastructures has increased importantly. Unfortunately, several times such progress is not directly translated into improved performance for all applications and users. Performance issues can vary from traffic congestion to packet loss caused by intermediate link failures. Such issues lead to financial losses for service providers and customers. Service level monitoring tools must be capable of detecting performance issues in an efficient manner. Measurement mechanisms are one of the most used tools employed for such monitoring. Service level monitoring is usually performed using either active or passive mechanisms. The network conditions are monitored in a non-intrusive way in passive mechanisms (e.g., IP Flow Information EXport - IPFIX [1]) since monitoring traffic is not created by the monitoring process itself. This is usually done through the observation of network flows J´ eferson C. Nobre is a former PhD candidate at the Federal University of Rio Grande do Sul, Brazil. He conducted this work during the Monitor- ing Program (Programa de Monitoramento) 2014, funded by the Brazilian National Research Network (Rede Nacional de Pesquisa). inside network devices. Active measurements (e.g., One-Way Active Measurement Protocol - OWAMP [2] and Cisco Service Level Assurance Protocol -IPSLA [3]), on the other hand, are intrusive because they inject monitoring traffic on the network infrastructure to deliver performance metrics. Active measurements can detect end-to-end performance issues in a fine grained fashion. Thus, service level monitoring is usually performed through active mechanisms. The activation of active measurement sessions is expen- sive in terms of the consumed resources (e.g., CPU cycles and memory footprint). Besides that, activated measurement sessions also increase the network load due to the injected traffic. At least, one session must be activated to measure a destination, thus the amount of consumed resources is a function of the number of measured destinations. Thus, with more possible paths the larger will be the resources needed to deploy such sessions. The best practice on the operation of active measurement mechanisms is to define the measure- ment sessions considering the expertise and knowledge of the network administrator. However, current SLA monitoring solutions consume a large amount of computational and human resources, which are not always available to the service provider. Measurement federations could help network operators to troubleshoot perceived abnormalities as well as improve network middleware regarding service level issues. In this context, active measurement mechanisms are one of the most important measurement tools in such federations. However, the sole employment of a measurement federation does not improve the capabilities of such mechanisms in important aspects, such as scalability and efficiency. For example, the current best practice regarding active measurements mecha- nisms usually covers only a fraction of the network flows that should be observed in order to save resources. This can lead to performance problems being missed. This practice is labor-intensive for the network administrator and inefficient considering highly dynamic network infrastructures since ad- ministrator reactions are necessary to reconfigure the active measurement mechanisms. In this context, novel approaches are needed to improve the operation of active measurement mechanisms in measurement federations. Measurement correlation has been used in order to enable efficient control loops for measurement sessions activation [4]. Despite the fact that measurement federations are com- 978-3-901882-89-0 @2017 IFIP 584