Illustration of ports, flows, host buffers and streams
This figure shows the naming conventions used for the data flow through the accelerator and into the host memory.
Frames from the network are received by physical accelerator ports. Accelerator ports are numbered from 0 to the number of physical accelerator ports minus one.
The frames from the ports are merged in arrival time order.
When the frames have passed through the frame decoder, hash and filter, frames are categorized into flows. Flows are groups of frames with the same properties with respect to filtering, port numbers and hash values.
The multi-CPU distribution functionality maps the flows to host buffers. From 1 to 128 host buffers can be defined with dynamic segment size, and from 1 to 64 host buffers can be defined with static segment size.
When data is transferred from the accelerator to the host buffers in user memory it is done via DMA.