• Commodity hardware
• Deep Packet Inspection
• DPI
• FastFlow
• Multiprocessor

In the last years we assisted to a large increase of the number of applications running on top of IP networks. Consequently the need to implement very efficient monitoring solutions that can manage these high data rates and that can classify the type of traffic which is traveling over the network has increased.

For example, as far as network security is concerned, in the recent years we have seen a shift from so-called "network-level" attacks, which target the network they are transported on (e.g. Denial of Service), to content-based threats which exploit applications vulnerabilities and require sophisticated levels of intelligence to be detected. For some of these threats, it is no more sufficient to have only a software solution on the client side but we also need to run some controls on the network itself. To manage these kinds of scenarios, payload inspection is often required in order to correctly identify the application protocol and to process the data carried over it. This is the reason why, in recent years, Deep Packet Inspection (DPI) technology has emerged.

This kind of processing is in many cases implemented, at least in part, through dedicated hardware. However, full software solutions may often be more appealing because they are typically more economical and have, in general, the capability to react faster to protocols evolution and changes.

Moreover, software solutions which run over general purpose hardware do not exploit the underlying multiprocessor architecture, providing only the capability to process the incoming packets sequentially. Furthermore, many DPI research works that can be found in literature and which exploits multicore architectures are often characterized by a poor scalability, due to the overhead required for synchronization and to load unbalance among the used cores.

In this thesis, we will describe the design and implementation of a DPI framework capable of managing current networks rates using commodity multicore hardware. Our framework provides the possibility to identify the protocol, to specify the kind of data to extract when it has been identified and how these data has to be processed.

Differently from existing works, the developed framework has been designed according to the structured parallel programming theory, allowing thus to completely hide to the user the complexity of the management of the problems related to an efficient exploitation of the underlying architecture. These concepts have then been applied using FastFlow, a library for structured parallel programming targeting both shared memory and distributed memory architectures.