Abstract
Data stream management systems (DSMSs) provide a new and alternative way to perceive and analyze data streams. Similar to the database management systems (DBMSs), the DSMSs use a declarative query language to handle data. One of the main differences is that the DSMSs obtain the data from streaming sources, for example a local area network (LAN), instead of a database. Such an approach opens up for a set of tasks that can be described using e.g. SQL-like queries.
To begin with, we discuss the networking application and introduce a set of requirements that might be useful for DSMSs in general. Some of these requirements are further discussed as we describe the issues in DSMSs. This thesis focuses on one particular DSMS, TelegraphCQ, and we give a thorough description and discussion of its features.
We have designed and implemented a set of tasks that may be of value for the network monitoring application as described in this thesis. We discuss these tasks, investigate their qualities and propose solutions on how to implement them in the declarative language provided by TelegraphCQ.
Finally, we run a performance analysis of some of the tasks to see how TelegraphCQ manages to handle data streams at varying loads. We focus on two metrics; relative throughput to the number of packets received, and accuracy of the results. These metrics are very important with respect to the reliability and applicability of TelegraphCQ. In this context, we implement an experiment setup for network monitoring with DSMSs, such that the results can be easily re-tested and verified. We show that TelegraphCQ only manages a network load of approximately 2.5 Mbits/s before it starts dropping packets.
We end the discussion by evaluating TelegraphCQ's support for the requirements described in the beginning of the thesis, and point out some of the requirements TelegraphCQ does not support. We discuss the results from the performance evaluation and conclude that the accuracy is satisfying. The conclusion is that, due to the low relative throughput, TelegraphCQ is not suited for network traffic monitoring at higher network loads.