Hardware-assisted high-speed connection tracking

Details

Supervisors

Barbette, Tom

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] en sciences informatiques, à finalité spécialisée

Abstract

The relentless growth in network traffic places a growing burden on server CPUs, diverting valuable resources and time that could be better utilized for higher-level application processing. In the middle of the packet processing operations is connection tracking, its goal is to associate the connection identifier of a packet to a connection state, the data that an application stores and uses to process all the packets belonging to a specific connection. Numerous researchers have dedicated efforts to enhance connection tracking, a pivotal aspect of network processing, by refining the conntrack mechanism. However, alongside the escalation in traffic, network interface cards (NICs) have also gained substantial processing power. Consequently, we have pursued the avenue of offloading the connection tracking process to the NIC, aiming to alleviate CPU strain and amplify overall network capabilities. We present a hybridized connection tracking solution that leverages the combined capabilities of both the CPU and NIC. This approach ensures optimal performance by avoiding any potential throttling of either component. We demonstrate that such hybridization is indeed feasible and can yield a straightforward 10\% increase in throughput. This improvement is achieved through simple traffic profiling mechanisms.