DPDK burst replay tool

Introduction

This is a modified version of the dpdk-replay tool, which was originally published on this GitHub repository. The tool is designed to provide high DPDK performances (up to 148Mpps with 64B packets) to burst any pcap dump on a single NIC port(s).

To do so, the pcap files will be cached on hugepages before being sent through DPDK.

How to play with it

Install dependencies

dpdk v23.11.2 (LTS) or later
libnuma-dev
libyaml-dev
libcyaml-dev
libcsv-dev
That's all.

$ sudo apt install libnuma-dev libyaml-dev libcyaml-dev libcsv-dev -y

NB: libpcap is not required, as dpdk-replay process pcap files manually.

Compiling and installing it (new way)

This is the recommended way to compile and install dpdk-replay.

Since the new dpdk-replay version relies on the libcyaml library, you need to make sure the repo is cloned with the --recurse-submodules option. If you haven't done it, you can still do it with the following command:

$ git submodule update --init --recursive

Then, you can compile and install dpdk-replay with the following commands:

$ mkdir build; cd build
$ cmake ..
$ make
$ sudo make install

Launching it (new way)

It is now possible to use a configuration file to configure the dpdk-replay tool.

The configuration file is a YAML file that contains the following syntax:

---
# Here you can specify the pcap file to replay
# and the number of tx queues
# Every trace will use a separate TX core to send packets
# If you want to push a specific trace to the limits (e.g., 148Mpps),
# you can specify the same trace N times, so that N cores will be used to
# send packets from the same trace.
# Of course, this will not ensure that the packets will be sent in the same order.
traces: 
  - path: "/users/smiano/maestro-eval/pcaps/uniform_64B.pcap"
    tx_queues: 8
# Specify here the numa node to use
numacore: 0
# Specify the number of runs to loop the pcap file
nbruns: 100000000
# If you specify the timeout, the PCAP will be replayed until the timeout is reached
timeout: 10
# Specify the maximum packets per second
max_mpps: -1
# Specify the maximum megabits per second
max_mbps: -1
# Set this to true if you want to write the results in a CSV file
write_csv: True
# Set this to true if you want to wait and press enter before starting the replay
wait_enter: False
# Set this to true if you want to enable slow mode
slow_mode: False
# Set this to true if you want to convert the results to JSON format (only if write_csv is True)
convert_to_json: True
# Specify the number of RX queues and cores
nb_rx_queues: 16
nb_rx_cores: 4
# Specify the PCI address of the NIC to use to send packets
send_port_pci: 0000:51:00.0
# Specify the PCI address of the NIC to use to read results from
# By default, the 1st PCI address of the send_port_pci should be in this list
# If you want to read results from multiple NICs, you can specify them here
stats:
  - pci_id: 0000:51:00.0
    file_name: "result_v1.csv"
  - pci_id: 0000:51:00.1
    file_name: "result_v2.csv"
# Specify the log level
loglevel: TRACE

Then, you can launch dpdk-replay with the following command:

$ dpdk-replay --config config.yaml

sebymiano / dpdk-burst-replay

readme