User plane

Besides signalling, rk sends and measures user-plane traffic over GTP-U, through the UPF under test.

How traffic starts

A flow starts traffic with the uplane_start action, usually once the PDU session is up:

it takes the tunnel from the session: the core's TEID and N3 address, the UE's IP and the QFI;
it sends for duration;
the flow continues when traffic completes, and the result is in the report's uplane_report.

uplane_start can override the session values:

Parameter	What it sets
`remote_teid`, `remote_addr`	the core's tunnel end
`local_teid`	the gNB's tunnel end (not zero)
`ue_ip`	the UE address
`target_addr`, `target_addr6`	where traffic goes; needed for an IPv4 UE
`duration`	how long to send
`qfi`	1–63; by default, the QFI the core granted
`protocol`	`icmp`, `udp` or `tcp`
`pps`	packets per second, 1–1,000,000
`payload_size`	1–65507 bytes, `USPACE` only
`df`	set the don't-fragment bit, `USPACE` only
`udp_dn`	`ul`, `dl` or `bidir`

Datapaths

The gNB's uplane.type picks how packets are sent:

Type	How	Build and licence
`USPACE`	kernel sockets	every build, no licence feature
`DPDK`	a DPDK helper process with its own NIC	the full build, Linux amd64, the licence feature `datapath.dpdk`, and enough `max_dpdk_cores` for `dpdk_lcores`
`MOCK`	sends nothing	for testing flows only; measures nothing

The community build has no DPDK code. Asking it for DPDK stops the run with exit code 4 and a message pointing to the full build.

USPACE tuning: gtpu_rcvbuf and gtpu_sndbuf (4 MiB by default; capped without CAP_NET_ADMIN), gtpu_readers, gtpu_writers, gtpu_batch (64), gtpu_no_gso, gtpu_no_gro.

DPDK needs these uplane fields:

mode and protocol;
lport_range_min and lport_range_max;
port_pci: a PCI address, or af_packet:<ifname>;
port_addr, port_gateway and target_addr;
listen_port_start or listen_port_num.

Optional: tx_burst (32 by default, 1–1024) and dpdk_lcores (1–64).

gNBs that share an N3 address share one socket. They must use the same type and protocol.

The far end: `rk server`

rk server answers the traffic rk sends, so you control both ends:

rk server uspace: a kernel-socket echo server. Needs -protocol; ICMP needs CAP_NET_RAW or root.
rk server dn: a data-network endpoint. With -gtpu, it terminates N3 itself on port 2152, with no core in between. -allow limits which UE addresses it answers (by default 10.45.0.0/16).
rk server dpdk: a DPDK reflector. Needs the full build and datapath.dpdk; -gtp and -gtpu-teid make it a GTP-U endpoint.

rk server -h lists each backend's flags.

How traffic starts

Datapaths

The far end: rk server

The far end: `rk server`