Mardi 6 octobre 2015 à 11h, salle 24-25/405
The analysis of massive data streams is fundamental in many monitoring applications (e.g, Internet routers). For networks operators, it is a recurrent and crucial issue to determine whether huge data streams, received at their monitored devices, are correlated or not as it may reveal the presence of attacks. First, we propose a metric, called codeviation, that allows to evaluate the correlation between distributed streams. This metric is inspired from classical metric in statistics and probability theory, and as such enables to understand how observed quantities change together, and in which proportion. We then propose to estimate the codeviation in the data stream model. In this model, functions are estimated on a huge sequence of data items, in an online fashion, and with a very small amount of memory with respect to both the size of the input stream and the values domain from which data items are drawn. We give upper and lower bounds on the quality of the codeviation, and provide both local and distributed algorithms that additively approximates the codeviation among data streams using sub-linear space. On the other hand, we consider the problem of identifying global iceberg attacks in massive and physically distributed streams. A global iceberg is a distributed denial of service attack, where some elements globally recur many times across the distributed streams, but locally, they do not appear as a deny of service. A natural solution to defend against global iceberg attacks is to rely on multiple routers that locally scan their network traffic, and regularly provide monitoring information to a server in charge of collecting and aggregating all the monitored information. Any relevant solution to this problem must minimise the communication between the routers and the coordinator, and the space required by each node to analyse its stream. We propose a distributed algorithm that tracks global icebergs on the fly with guaranteed error bounds, limited memory and processing requirements. We present a thorough analysis of our algorithm performance. In particular we derive an optimal upper bound on the number of bits communicated between the multiple routers and the coordinator in presence of an oblivious adversary. Finally, we present the main results of the experiments we have run on a cluster of single-board computers. Those experiments confirm the efficiency and accuracy of our algorithm to track global icebergs hidden in very large input data streams exhibiting different shapes.