A Low Power, Programmable Networking Platform and Development Environment - Advances in Network-Embedded Management and Applications - page 33

The C-Board with the DPI firmware architecture is capable of monitoring
full speed the 10Gbps traffic in both directions, and it can forward it to the
SFP interfaces without packet loss.
Finally, in a realistic use-case a part of the traffic can be streamed to the
SFPs for packet level analysis. Up to 8 Gbps traffic can be streamed to the
receiving monitor PCs dependably [13].

5.2 Generic Switch/Router architecture

The SCALOPES C-Board can also be used as a generic Switching/Routing
architecture with hardware level packet forwarding. The architecture supports
two 10 GbE ports, 16 GbE ports with an arbitrary combination of SFPs.

Fig . 6. Generic Packet forwarding architecture

The used packet processing pipeline is similar to the model recommended
by Xilinx [15] and the model used by the Liberouter project [5]. The packets
are filtered first, then all traffic is written in the DDRAM for buffering. Since
the memory access can be the bottleneck in our packet processing pipeline,
we tried to avoid copying the packet. We have decided to use the shared-
memory packet forwarding model, this way avoiding unnecessary copying of
data. The model has a further advantage: even multicasting/broadcasting can
be done without actually copying the data.
The generic packet forwarding architecture is shown by Fig. 6 . The arriving
packets are buffered, while their header information is however forwarded to
the lookup module. The lookup module is responsible to decide on which
interface the packet should leave. For flooding, broadcast and multicast
multiple interfaces may be selected. For different application types, different
routing/switching modules can be designed, which may operate at different
layers of the OSI model, like L2 switching, IPv4 or IPv6 routing, MPLS
packet switching etc. Based on the decision at the lookup module, a short