Packet Filtering Module For PFQ Packet Capturing Engine.

The evolution of commodity hardware is pushing parallelism forward as the key factor that can allow software to attain hardware-class performance while still retaining its advantages. On one side, commodity CPUs are providing more and more cores (the next-generation Intel Xeon E 7500 CPUs will soon make 10 cores processors a commodity product), with a complex cache hierarchy which makes aware data placement crucial to good performance. On the other side, server NIC‘s are adapting to these new trends by increasing themselves their level of parallelism. While traditional 1Gbps NICs exchanged data with the CPU through a single ring of shared memory buffers, modern 10Gbps cards support multiple queues: multiple cores can therefore receive and transmit packets in parallel. In particular, incoming packets can be de-multiplexed across CPUs based on a hash function (the so-called RSS technology) or on the MAC address (the VMD-q technology, designed for servers hosting multiple virtual machines). The Linux kernel has recently begun to support these new technologies. Though there is lot of network monitoring software‘s, most of them have not yet been designed with high parallelism in mind. Therefore a novel packet capturing engine, named PFQ was designed, that allows efficient capturing and in-kernel aggregation, as well as connection-aware load balancing. Such an engine is based on a novel lockless queue and allows parallel packet capturing to let the user-space application arbitrarily define its degree of parallelism. Therefore, both legacy applications and natively parallel ones can benefit from such capturing engine. In addition, PFQ outperforms its competitors both in terms of captured packets and CPU consumption. In this thesis, a new packet filtering block is designed implemented and added to the existing PFQ capture engine which helps in dropping out unnecessary packets before they are copied into the kernel space thus improves the overall performance of the engine considerably. Because network monitors often want only a small subset of network traffic, a dramatic performance gain is realized by filtering out unwanted packets in interrupt context