This is a simple description of the procedure to install the traffic generator TRex. The detailed documentation for TRex can be found here. This is intended as a preparation for measuring the performance of open source UPFs.
- Simple Overview of TRex and DUT (UPF)
- Install TRex
- Setup TRex
- Run TRex
- How to capture packets on DPDK ports
- Sample Configurations
- Changelog (summary)
This describes a simple configuration of TRex and DUT (UPF). Note that this configuration is implemented with Proxmox VE VMs.
The following minimum configuration was set as a condition.
- One TRex and DUT (UPF)
The built simulation environment is as follows.
The TRex used is as follows.
- TRex v3.08 (2025.11.05) - https://github.com/cisco-system-traffic-generator/trex-core
- Scapy v2.6.1 (2024.11.05) - https://github.com/secdev/scapy
Each VMs are as follows.
| VM | SW & Role | IP address | OS | CPU (Min) |
Mem (Min) |
HDD (Min) |
|---|---|---|---|---|---|---|
| VM-TG | TRex Traffic Generator |
192.168.0.131/24 | Ubuntu 24.04 | 3 | 8GB | 10GB |
| VM-DUT | each UPF DUT (Device Under Test) |
192.168.0.151/24 | Ubuntu 24.04 | 2 | 8GB | 10GB |
The network interfaces of each VM are as follows.
| VM | Device | Model | Linux Bridge | IP address | Interface | Under DPDK |
|---|---|---|---|---|---|---|
| VM-TG | -- | |||||
| ens19 | VirtIO | mgbr0 | 192.168.0.131/24 | (Mgmt NW) | -- | |
| ens20 | VirtIO | vmbr3 | 192.168.13.131/24 | N3 | x | |
| ens21 | VirtIO | vmbr6 | 192.168.16.152/24 | N6 | x | |
| VM-DUT | -- | |||||
| ens19 | VirtIO | mgbr0 | 192.168.0.151/24 | (Mgmt NW) | -- | |
| ens20 | VirtIO | vmbr3 | 192.168.13.151/24 | N3 | -- | |
| ens21 | VirtIO | vmbr4 | 192.168.14.151/24 | N4 | -- | |
| ens22 | VirtIO | vmbr6 | 192.168.16.151/24 | N6 | -- |
Linux Bridges of Proxmox VE are as follows.
| Linux Bridge | Network CIDR | Interface |
|---|---|---|
| vmbr1 | 10.0.0.0/24 | NAPT NW |
| mgbr0 | 192.168.0.0/24 | Mgmt NW |
| vmbr3 | 192.168.13.0/24 | N3 |
| vmbr4 | 192.168.14.0/24 | N4 |
| vmbr6 | 192.168.16.0/24 | N6 |
UE IP address and TEID are as follows.
| UE IP address | UpLink TEID | DownLink TEID |
|---|---|---|
| 10.45.0.2/24 | 0x00000001 | 0x00000002 |
Please refer to the following for installing TRex.
- TRex v3.08 (2025.11.05) - https://github.com/cisco-system-traffic-generator/trex-core/wiki
This section explains how to install TRex v3.08 to /opt/trex directory.
This time, for using the PDU Session container in the GTP-U packet header, I will replace gtp.py and gtp_v2.py of Scapy v2.4.3 included in TRex v3.08 with those of Scapy v2.6.1.
First, download the pre-built TRex v3.08 binaries and extract it to /opt/trex directory.
# cd /opt
# wget --no-check-certificate https://trex-tgn.cisco.com/trex/release/v3.08.tar.gz
# tar xfvz v3.08.tar.gz
# mv v3.08 trex
Then, for using the PDU Session container in the GTP-U packet header, replace gtp.py and gtp_v2.py of Scapy v2.4.3 included in TRex v3.08 with those of Scapy v2.6.1.
# wget https://raw.githubusercontent.com/secdev/scapy/refs/tags/v2.6.1/scapy/contrib/gtp.py -O /opt/trex/external_libs/scapy-2.4.3/scapy/contrib/gtp.py
# wget https://raw.githubusercontent.com/secdev/scapy/refs/tags/v2.6.1/scapy/contrib/gtp_v2.py -O /opt/trex/external_libs/scapy-2.4.3/scapy/contrib/gtp_v2.py
First, check the network devices and bus information of the installation machine.
# lshw -c network -businfo
Bus info Device Class Description
====================================================
pci@0000:00:12.0 network Virtio network device
virtio@1 ens18 network Ethernet interface
pci@0000:00:13.0 network Virtio network device
virtio@2 ens19 network Ethernet interface
pci@0000:00:14.0 network Virtio network device
virtio@3 ens20 network Ethernet interface
pci@0000:00:15.0 network Virtio network device
virtio@4 ens21 network Ethernet interface
In my environment, based on this information, configure pci@0000:00:14.0 of ens20(N3) and pci@0000:00:15.0 of ens21(N6).
Use the following dpdk_setup_ports.py script to create /etc/trex_cfg.yaml with the IP settings for the ports used by DPDK.
# cd /opt/trex
# ./dpdk_setup_ports.py -i
...
I edited this file created as follows.
/etc/trex_cfg.yaml
### Config file generated by dpdk_setup_ports.py ###
- version: 2
interfaces: ['00:14.0', '00:15.0']
port_info:
- ip: 192.168.13.131
default_gw: 192.168.13.151
- ip: 192.168.16.152
default_gw: 192.168.16.151
platform:
master_thread_id: 0
latency_thread_id: 1
dual_if:
- socket: 0
threads: [2]
I am using the following for the TRex load profile. Also, the payload size is set to 1400 bytes and the QFI is set to 1.
Create gtp_1pkt_simple.py for UpLink load profile in /opt/trex/stl directory and set the following parameters.
| Item | Value (my environment) |
|---|---|
| GNB_IP_V4 | "192.168.13.131" |
| N3_IP_V4 | "192.168.13.151" |
| UE_IP_V4 | "10.45.0.2" |
| DN_IP_V4 | "192.168.16.152" |
| UL_TEID | 0x00000001 |
/opt/trex/stl/gtp_1pkt_simple.py
from trex_stl_lib.api import *
from scapy.contrib.gtp import GTP_U_Header, GTPPDUSessionContainer
import argparse
GNB_IP_V4 = "192.168.13.131"
N3_IP_V4 = "192.168.13.151"
UE_IP_V4 = "10.45.0.2"
DN_IP_V4 = "192.168.16.152"
UL_TEID = 0x00000001
class STLS1(object):
def create_stream (self):
return STLStream(
packet =
STLPktBuilder(
pkt = Ether()/IP(src=GNB_IP_V4,dst=N3_IP_V4,version=4)/
UDP(dport=2152,sport=2152)/
GTP_U_Header(teid=UL_TEID)/
GTPPDUSessionContainer(type=1,QFI=1)/
IP(src=UE_IP_V4,dst=DN_IP_V4,version=4)/
UDP(dport=1234,sport=1234)/
(1400*'x')
),
mode = STLTXCont())
def get_streams (self, tunables, **kwargs):
parser = argparse.ArgumentParser(description='Argparser for {}'.format(os.path.basename(__file__)),
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
args = parser.parse_args(tunables)
# create 1 stream
return [ self.create_stream() ]
def register():
return STLS1()Create udp_1pkt_simple.py for DownLink load profile in /opt/trex/stl directory and set the following parameters.
| Item | Value (my environment) |
|---|---|
| UE_IP_V4 | "10.45.0.2" |
| DN_IP_V4 | "192.168.16.152" |
/opt/trex/stl/udp_1pkt_simple.py
from trex_stl_lib.api import *
import argparse
UE_IP_V4 = "10.45.0.2"
DN_IP_V4 = "192.168.16.152"
class STLS1(object):
def create_stream (self):
return STLStream(
packet =
STLPktBuilder(
pkt = Ether()/IP(src=DN_IP_V4,dst=UE_IP_V4,version=4)/
UDP(dport=1234,sport=1234)/
(1400*'x')
),
mode = STLTXCont())
def get_streams (self, tunables, **kwargs):
parser = argparse.ArgumentParser(description='Argparser for {}'.format(os.path.basename(__file__)),
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
args = parser.parse_args(tunables)
# create 1 stream
return [ self.create_stream() ]
def register():
return STLS1()I am using the following for the TRex latency profile. Also, the payload size is set to 1400 bytes and the QFI is set to 1.
Note. In TRex STL mode, using STLFlowLatencyStats on a UDP packet will invalidate the UDP checksum value. The reason is that when using STLFlowLatencyStats, the last 16 bytes of the payload are rewritten. This will result in an incorrect UDP checksum value on the receiving side. Therefore, set the UDP checksum value to zero to ignore it.
And for perfomance reasons, latency streams are not affected by -m flag when running start command in the trex console, so you can only change the pps value by editing the profile code. Here, STLTXCont(pps=1) is set in mode argument of STLStream() as follows.
mode = STLTXCont(pps=1)
Create gtp_latency_1pkt_simple.py for UpLink latency profile in /opt/trex/stl directory and set the following parameters.
| Item | Value (my environment) |
|---|---|
| GNB_IP_V4 | "192.168.13.131" |
| N3_IP_V4 | "192.168.13.151" |
| UE_IP_V4 | "10.45.0.2" |
| DN_IP_V4 | "192.168.16.152" |
| UL_TEID | 0x00000001 |
/opt/trex/stl/gtp_latency_1pkt_simple.py
from trex_stl_lib.api import *
from scapy.contrib.gtp import GTP_U_Header, GTPPDUSessionContainer
import argparse
GNB_IP_V4 = "192.168.13.131"
N3_IP_V4 = "192.168.13.151"
UE_IP_V4 = "10.45.0.2"
DN_IP_V4 = "192.168.16.152"
UL_TEID = 0x00000001
class STLS1(object):
def create_stream (self):
return STLStream(
packet =
STLPktBuilder(
pkt = Ether()/IP(src=GNB_IP_V4,dst=N3_IP_V4,version=4)/
UDP(dport=2152,sport=2152,chksum=0)/
GTP_U_Header(teid=UL_TEID)/
GTPPDUSessionContainer(type=1,QFI=1)/
IP(src=UE_IP_V4,dst=DN_IP_V4,version=4)/
UDP(dport=1234,sport=1234)/
(1400*'x')
),
flow_stats = STLFlowLatencyStats(pg_id=0),
mode = STLTXCont(pps=1))
def get_streams (self, tunables, **kwargs):
parser = argparse.ArgumentParser(description='Argparser for {}'.format(os.path.basename(__file__)),
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
args = parser.parse_args(tunables)
# create 1 stream
return [ self.create_stream() ]
def register():
return STLS1()Create udp_latency_1pkt_simple.py for DownLink latency profile in /opt/trex/stl directory and set the following parameters.
| Item | Value (my environment) |
|---|---|
| UE_IP_V4 | "10.45.0.2" |
| DN_IP_V4 | "192.168.16.152" |
/opt/trex/stl/udp_latency_1pkt_simple.py
from trex_stl_lib.api import *
import argparse
UE_IP_V4 = "10.45.0.2"
DN_IP_V4 = "192.168.16.152"
class STLS1(object):
def create_stream (self):
return STLStream(
packet =
STLPktBuilder(
pkt = Ether()/IP(src=DN_IP_V4,dst=UE_IP_V4,version=4)/
UDP(dport=1234,sport=1234,chksum=0)/
(1400*'x')
),
flow_stats = STLFlowLatencyStats(pg_id=1),
mode = STLTXCont(pps=1))
def get_streams (self, tunables, **kwargs):
parser = argparse.ArgumentParser(description='Argparser for {}'.format(os.path.basename(__file__)),
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
args = parser.parse_args(tunables)
# create 1 stream
return [ self.create_stream() ]
def register():
return STLS1()Set vm.nr_hugepages to 1024.
# sysctl vm.nr_hugepages=1024
Then mount the HugePages as hugetlbfs(HugeTLB) file system.
# mkdir -p /mnt/huge
# mount -t hugetlbfs nodev /mnt/huge
Also, down ens18 interface on VM-TG and disable the default GW.
# ip link set dev ens18 down
Please configure and start the DUT (UPF in my environment) in advance.
For information on how to use the TRex console, see here.
First, start the TRex server.
# cd /opt/trex
# ./t-rex-64 -i --no-scapy-server
Then, open another terminal and connect to the TRex server to check the status of each port.
# cd /opt/trex
# ./trex-console
...
trex>
trex>portattr
Port Status
port | 0 | 1
----------------+----------------------+---------------------
driver | net_virtio | net_virtio
description | Virtio network dev | Virtio network dev
link status | UP | UP
link speed | 200 Gb/s | 200 Gb/s
port status | IDLE | IDLE
promiscuous | off | off
multicast | off | off
flow ctrl | N/A | N/A
vxlan fs | - | -
-- | |
layer mode | IPv4 | IPv4
src IPv4 | 192.168.13.131 | 192.168.16.152
IPv6 | off | off
src MAC | bc:24:11:51:b6:55 | bc:24:11:18:b4:63
--- | |
Destination | 192.168.13.151 | 192.168.16.151
ARP Resolution | bc:24:11:4a:0c:dc | bc:24:11:42:ed:8e
---- | |
VLAN | - | -
----- | |
PCI Address | 0000:00:14.0 | 0000:00:15.0
NUMA Node | -1 | -1
RX Filter Mode | hardware match | hardware match
RX Queueing | off | off
Grat ARP | every 120 seconds | every 120 seconds
------ | |
trex>
Switch ports: [0, 1] into service mode and ping to the IP addresses (192.168.13.151 and 192.168.16.151) of the DUT.
trex>service
Enabling service mode on port(s): [0, 1] [SUCCESS]
10.62 [ms]
trex(service)>
trex(service)>ping -p 0 -d 192.168.13.151 -n 3
Pinging 192.168.13.151 from port 0 with 64 bytes of data:
Reply from 192.168.13.151: bytes=64, time=2.00ms, TTL=64
Reply from 192.168.13.151: bytes=64, time=1.26ms, TTL=64
Reply from 192.168.13.151: bytes=64, time=2.61ms, TTL=64
trex(service)>
trex(service)>ping -p 1 -d 192.168.16.151 -n 3
Pinging 192.168.16.151 from port 1 with 64 bytes of data:
Reply from 192.168.16.151: bytes=64, time=33.58ms, TTL=64
Reply from 192.168.16.151: bytes=64, time=1.84ms, TTL=64
Reply from 192.168.16.151: bytes=64, time=3.05ms, TTL=64
trex(service)>
Once confirmed that ping works, the service mode exit.
trex(service)>service --off
Disabling service mode on port(s): [0, 1] [SUCCESS]
7.59 [ms]
trex>
Then, apply traffic to the DUT.
In the following example, use the load profile gtp_1pkt_simple.py to apply GTP-U traffic to the DUT at 150 Kpps for 60 seconds via port 0.
trex>start -f stl/gtp_1pkt_simple.py -p 0 -m 150kpps -d 60
To check the traffic statistics, type tui in the TRex console to switch the view.
trex>tui
Below are some sample statistics. According to this, 626.15 Mbps of 1.79 Gbps was dropped, and 1.16 Gbps was received.
Global Statistics
connection : localhost, Port 4501 total_tx_L2 : 1.79 Gbps
version : STL @ v3.08 total_tx_L1 : 1.81 Gbps
cpu_util. : 1.81% @ 1 cores (1 per dual port) total_rx : 1.16 Gbps
rx_cpu_util. : 0.18% / 100.57 Kpps total_pps : 150.13 Kpps
async_util. : 0% / 10.01 bps drop_rate : 626.15 Mbps
total_cps. : 0 cps queue_full : 15,046 pkts
Port Statistics
port | 0 | 1 | total
-----------+-------------------+-------------------+------------------
owner | root | root |
link | UP | UP |
state | TRANSMITTING | IDLE |
speed | 200 Gb/s | 200 Gb/s |
CPU util. | 1.81% | 0.0% |
-- | | |
Tx bps L2 | 1.79 Gbps | 0 bps | 1.79 Gbps
Tx bps L1 | 1.81 Gbps | 0 bps | 1.81 Gbps
Tx pps | 150.13 Kpps | 0 pps | 150.13 Kpps
Line Util. | 0.91 % | 0 % |
--- | | |
Rx bps | 0 bps | 1.16 Gbps | 1.16 Gbps
Rx pps | 0 pps | 100.57 Kpps | 100.57 Kpps
---- | | |
opackets | 3727565 | 1 | 3727566
ipackets | 0 | 2451753 | 2451753
obytes | 5554071850 | 46 | 5554071896
ibytes | 0 | 3545233438 | 3545233438
tx-pkts | 3.73 Mpkts | 1 pkts | 3.73 Mpkts
rx-pkts | 0 pkts | 2.45 Mpkts | 2.45 Mpkts
tx-bytes | 5.55 GB | 46 B | 5.55 GB
rx-bytes | 0 B | 3.55 GB | 3.55 GB
----- | | |
oerrors | 0 | 0 | 0
ierrors | 0 | 0 | 0
status: \
Press 'ESC' for navigation panel...
status:
tui>
In the following example, use the latency profile gtp_latency_1pkt_simple.py to apply GTP-U traffic to the DUT at 1 pps for 10 seconds via port 0.
trex>start -f stl/gtp_latency_1pkt_simple.py -p 0 -d 10
To check the latency statistics, type stats -l in the TRex console.
trex>stats -l
Below are some sample statistics. According to this, The average latency is 0.285 (msec), the maximum is 0.307 (msec), and the minimum is 0.256 (msec).
Latency Statistics
PG ID | 0
-------------+---------------
TX pkts | 11
RX pkts | 11
Max latency | 307
Min latency | 256
Avg latency | 285
-- Window -- |
Last max | 279
Last-1 |
Last-2 |
Last-3 |
Last-4 |
Last-5 |
Last-6 |
Last-7 |
Last-8 |
Last-9 |
Last-10 |
Last-11 |
Last-12 |
Last-13 |
--- |
Jitter | 20
---- |
Errors | 0
trex>
In the following example, use the load profile udp_1pkt_simple.py to apply UDP traffic to the DUT at 150 Kpps for 60 seconds via port 1.
trex>start -f stl/udp_1pkt_simple.py -p 1 -m 150kpps -d 60
To check the traffic statistics, type tui in the TRex console to switch the view.
trex>tui
Below are some sample statistics. According to this, 671.65 Mbps of 1.73 Gbps was dropped, and 1.06 Gbps was received.
Global Statistics
connection : localhost, Port 4501 total_tx_L2 : 1.73 Gbps
version : STL @ v3.08 total_tx_L1 : 1.75 Gbps
cpu_util. : 2.36% @ 1 cores (1 per dual port) total_rx : 1.06 Gbps
rx_cpu_util. : 0.06% / 89.04 Kpps total_pps : 149.32 Kpps
async_util. : 0% / 8.26 bps drop_rate : 671.65 Mbps
total_cps. : 0 cps queue_full : 67,608 pkts
Port Statistics
port | 0 | 1 | total
-----------+-------------------+-------------------+------------------
owner | root | root |
link | UP | UP |
state | IDLE | TRANSMITTING |
speed | 200 Gb/s | 200 Gb/s |
CPU util. | 0.0% | 2.36% |
-- | | |
Tx bps L2 | 0 bps | 1.73 Gbps | 1.73 Gbps
Tx bps L1 | 0 bps | 1.75 Gbps | 1.75 Gbps
Tx pps | 0 pps | 149.32 Kpps | 149.32 Kpps
Line Util. | 0 % | 0.88 % |
--- | | |
Rx bps | 1.06 Gbps | 0 bps | 1.06 Gbps
Rx pps | 89.04 Kpps | 0 pps | 89.04 Kpps
---- | | |
opackets | 1 | 6083347 | 6083348
ipackets | 3453893 | 0 | 3453893
obytes | 46 | 8796519762 | 8796519808
ibytes | 5118667990 | 0 | 5118667990
tx-pkts | 1 pkts | 6.08 Mpkts | 6.08 Mpkts
rx-pkts | 3.45 Mpkts | 0 pkts | 3.45 Mpkts
tx-bytes | 46 B | 8.8 GB | 8.8 GB
rx-bytes | 5.12 GB | 0 B | 5.12 GB
----- | | |
oerrors | 0 | 0 | 0
ierrors | 0 | 0 | 0
status: -
Press 'ESC' for navigation panel...
status:
tui>
In the following example, use the latency profile udp_latency_1pkt_simple.py to apply UDP traffic to the DUT at 1 pps for 10 seconds via port 1.
trex>start -f stl/udp_latency_1pkt_simple.py -p 1 -d 10
To check the latency statistics, type stats -l in the TRex console.
trex>stats -l
Below are some sample statistics. According to this, The average latency is 0.314 (msec), the maximum is 0.368 (msec), and the minimum is 0.262 (msec).
Latency Statistics
PG ID | 1
-------------+---------------
TX pkts | 11
RX pkts | 11
Max latency | 368
Min latency | 262
Avg latency | 314
-- Window -- |
Last max | 285
Last-1 |
Last-2 |
Last-3 |
Last-4 |
Last-5 |
Last-6 |
Last-7 |
Last-8 |
Last-9 |
Last-10 |
Last-11 |
Last-12 |
Last-13 |
--- |
Jitter | 26
---- |
Errors | 0
trex>
There are three ways to do this.
- How to use packet capturing feature of TRex
- How to run
dpdk-dumpcap - How to run
tcpdumportsharkon another VM by configuring a bridge interface linked to a network interface under DPDK control
With the above steps, TRex has been constructed. Although the settings differ for each UPF, I believe this has made some progress in preparing to measure open source UPFs performance using TRex. I would like to thank the excellent developers and all the contributors of TRex.
- Simple PFCP Client
- Simple Measurement of UPF Performance 6
- Simple Measurement of UPF Performance 9
- Simple Measurement of UPF Performance 10
- [2025.11.29] Updated TRex from v3.06 to v3.08 and the OS from Ubuntu 22.04 to 24.04.
- [2025.06.09] Added the latency measurement.
- [2025.01.31] Changed the installation method to update
gtp.pyandgtp_v2.pyfor using the PDU Session container with the pre-built TRex v3.06 binaries. - [2025.01.16] Added the downlink measurement.
- [2024.12.08] Updated Scapy from v2.4.3 to v2.6.1 and built TRex for using the PDU Session container in the GTP-U packet header. The reason is that a proper QFI in the PDU Session container is now required when sending GTP-U packets to Open5GS UPF with the result of this fix. Please refer to here for details.
- [2024.11.03] Initial release.