PFQ Kernel Module
PFQ is a Linux Kernel module for packet capture and in-kernel functional processing designed to be efficient on multi-core architectures.
PFQ works by registering itself as a network sniffer (as PF_PACKET), or through accelerated drivers obtained by recompiling the source codes by pfq-omatic script.
description: Functional Networking Framework for Multi-core Architectures
author: Nicola Bonelli <nicola@pfq.io>
license: GPL
depends: crc16
vermagic: 3.16.0-4-amd64 SMP mod_unload modversions
parm: capture_incoming: Capture incoming packets: (1 default) (int)
parm: capture_outgoing: Capture outgoing packets: (0 default) (int)
parm: capt_slot_size: Maximum capture length (bytes) (int)
parm: xmit_slot_size: Maximum transmission length (default=1514 bytes) (int)
parm: capt_batch_len: Capture batch queue length (int)
parm: xmit_batch_len: Transmit batch queue length (int)
parm: vl_untag: Enable vlan untagging (default=0) (int)
parm: skb_pool_size: Socket buffer pool size (default=1024) (int)
parm: tx_affinity: Tx threads cpus' affinity (array of int)
PFQ provides similar performance to that of PF_RING-DNA and Netmap, while maintaining a total flexible design and the full compatibility with a plethora of network device drivers.
The software acceleration is achieved with a lock-free architecture and by overloading certain kernel functions at compile with more efficient implementations. Additional information is available in the section pfq-omatic.
PFQ takes advantage from:
- A total lock-free architecture on the fast-data path.
- Efficient implementations of queues (MPSC, SPMC) with amortized atomic operations.
- Reduced inter-cache traffic (cache coherence) among cores.
- Ad-hoc socket buffer memory allocator.
- Shared memory between user and kernel space on-top-of HugePages.
- Lightweight sparse counters.
- No false-sharing.