example.yml

# path of source code
local-workspace: /path/to/your/lumina/project/lumina/

# path of results
result-path: /path/to/your/results/

# file name of aggregated pcap
aggregate-pcap-filename: merge.pcap

# rewrite udp dst port for roce packets (4791-->9741)
rewrite-udp-dst-port: 9741

# number of times to run this experiment
num-repeats: 5

# switch config
switch:
  sde-path: /path/to/bf-sde/on/switch/ # the path where bf sde is installed
  workspace: /path/to/lumina/code/on/switch/ # the path where the lumina code and results are placed
  control-ip: 192.168.1.11 # the control ip of switch
  username: user # username of switch
  ssh-key-filepath: /home/foobar/.ssh/id_rsa # the private key path; the public key should be put to the remote side's authorized_keys
  port-speed: 100G # port speed for all the ports involved in the test
  fec-type: RS # fec type for all the ports involved in the test
  listen-port: 12345 # The port to listen (during the test, communicating with the traffic generator)
  local-switch-config-path: /path/to/your/switch-config.yml # Lumina will generate a switch config file there

# server configs
requester:
  workspace: /path/to/lumina/code/on/server/ # the path where the lumina code and results are placed
  control-ip: 192.168.1.12 # the control ip of requester
  username: user # username of requester
  ssh-key-filepath: /home/foobar/.ssh/id_rsa # the private key path; the public key should be put to the remote side's authorized_keys
  nic:
    type: CX6 # The NIC type to test; we support: CX4, CX5, CX6, E810
    if-name: ens1 # The interface name of the NIC
    pci-addr: "03:00.0" # The pcie address of the NIC
    ib-device: mlx5_0 # The ib device name of the RNIC
    ib-port: 1 # The ib port
    switch-port: 132 # The port number of the switch port that this NIC is connected to
    ip-list: # The ip addresses on the NIC
      - 10.0.0.1/24
      - 10.0.0.11/24
      - 10.0.0.21/24
      - 10.0.0.31/24
  roce-parameters: # RoCE parameters
    dcqcn-rp-enable: False
    dcqcn-np-enable: True
    min-time-between-cnps: 0
    adaptive-retrans: False
    slow-restart: True

responder:
  workspace: /path/to/lumina/code/on/server/
  control-ip: 192.168.1.13
  username: user
  ssh-key-filepath: /home/foobar/.ssh/id_rsa
  nic:
    type: CX6
    if-name: ens1
    pci-addr: "03:00.0"
    ib-device: mlx5_0
    ib-port: 1
    switch-port: 140
    ip-list:
      - 10.0.0.2/24
      - 10.0.0.12/24
      - 10.0.0.22/24
      - 10.0.0.32/24
  roce-parameters:
    dcqcn-rp-enable: True
    dcqcn-np-enable: True
    min-time-between-cnps: 0
    adaptive-retrans: False
    slow-restart: True

requester-mirror:
  workspace: /path/to/lumina/code/on/mirror-server/
  control-ip: 192.168.1.14
  username: user
  ssh-key-filepath: /home/foobar/.ssh/id_rsa
  pkt-dump-conf: # Config for packet dump program
    dump-filename: requester.pcap # The pcap filename
    snap-len: 128 # The maximum packet length for each packet (We are interested in the headers only)
    num-pkts: 16383 # The maximum number of packets to capture
  nic:
    if-name: ens1
    pci-addr: "03:00.0"
    switch-port: 148
    rte-port-id: 0

responder-mirror:
  workspace: /path/to/lumina/code/on/mirror-server/
  control-ip: 192.168.1.15
  username: user
  ssh-key-filepath: /home/foobar/.ssh/id_rsa
  pkt-dump-conf:
    dump-filename: responder.pcap
    snap-len: 128
    num-pkts: 16383
  nic:
    if-name: ens1
    pci-addr: "03:00.0"
    switch-port: 156
    rte-port-id: 0

traffic:
  num-qps: 4 # Number of connections
  rdma-verb: write # The rdma verb of the traffic
  num-msgs-per-qp: 1 # Number of messages to send per connection
  mtu: 1024 # MTU
  message-size: 1024000 # Size of each message
  listen-port: 13579 # The port 
  multi-gid: true # Whether to enable multi GID
  barrier-sync: false # Whether to enable barrier synchronization between the connections
  tx-depth: 1 # Tx depth
  min-retransmit-timeout: 14 # Min retransmission delay for timeout retransmission; the actual time calculation is 4.096*2^min-retransmit-timeout(check spec for detailed descripsion) 
  max-retransmit-retry: 7 # Max number of try for timeout retransmission
  data-pkt-events: # A list of events to inject
    - qpn:  0   # The (relative) QP number to inject event; the number starts from 0
      psn:  4   # The (relative) packet sequence number of the target packet; the number starts from 0
      type: ecn # The type of event
      iter: 1   # The "iteration" number of the target packet; maintained by switch (check our paper for more details); the number starts from 1
    - qpn:  1
      psn:  4
      type: ecn
      iter: 1
    - qpn:  2
      psn:  3
      type: drop
      iter: 1
    - qpn:  3
      psn:  5
      type: drop
      iter: 1