NP hardware acceleration generally alters packet flow as follows:
- Packets initiating a session pass to the FortiGate unit’s main processing resources (CPU).
- The FortiGate unit assesses whether the session matches fast path (offload) requirements.
To be suitable for offloading, traffic must possess only characteristics that can be processed by the fast path. The list of requirements depends on the processor, see NP6 session fast path requirements.
If the session can be fast pathed, the FortiGate unit sends the session key or IPsec security association (SA) and configured firewall processing action to the appropriate network processor.
- Network processors continuously match packets arriving on their attached ports against the session keys and SAs they have received.
- If a network processor’s network interface is configured to perform hardware accelerated anomaly checks, the network processor drops or accepts packets that match the configured anomaly patterns. These checks are separate from and in advance of anomaly checks performed by IPS, which is not compatible with network processor offloading. See .
- The network processor next checks for a matching session key or SA. If a matching session key or SA is found, and if the packet meets packet requirements, the network processor processes the packet according to the configured action and then sends the resulting packet. This is the actual offloading step. Performing this processing on the NP processor improves overall performance because the NP processor is optimized for this task. As well, overall FortiGate performance is improved because the CPU has fewer sessions to process.
- If a matching session key or SA is not found, or if the packet does not meet packet requirements, the packet cannot be offloaded. The network processor sends the data to the FortiGate unit’s CPU, which processes the packet.
Encryption and decryption of IPsec traffic originating from the FortiGate can utilize network processor encryption capabilities.
Packet forwarding rates vary by the percentage of offloadable processing and the type of network processing required by your configuration, but are independent of frame size. For optimal traffic types, network throughput can equal wire speed.