A profile with fleet.count: N spawns N independent simulated CPEs in one process. Each gets its own parameter tree, transport (cookie jar / Digest auth cache), event tracker, session, generator runner, and stamped per-instance values. Bootstrap Informs fire in parallel.

Minimal fleet¶

fleet:
  count: 100
  serialPattern: "TEST-{i:04}"

Each CPE's Device.DeviceInfo.SerialNumber (the leaf named in deviceIdPaths.serialNumber) gets stamped with the pattern: TEST-0001, TEST-0002, ..., TEST-0100.

{i} is the 1-based instance index. {i:N} is the same with zero-padding to N digits. {base} substitutes the literal value: from the SerialNumber leaf.

Per-CPE differentiation¶

Real fleets need more than just unique serials. WAN IPs, MAC addresses, hostnames, SSIDs all differ per device. Two mechanisms:

Inline placeholders (one-off)¶

Recognized in any leaf value when fleet.count > 1:

Named pools (referenced from many leaves)¶

Declare CIDRs once in fleet.pools, reference them by name:

fleet:
  count: 1000
  serialPattern: "TEST-{i:04}"
  pools:
    wan_ipv4:
      type: ipv4
      cidr: "203.0.113.0/24"           # RFC 5737 docs prefix; capacity 254
    wan_ipv6:
      type: ipv6
      cidr: "2001:db8:1::/64"           # RFC 3849 docs prefix
    delegated_prefix:
      type: ipv6prefix
      super: "2001:db8:cafe::/48"       # operator-side super-prefix
      sublen: 56                         # /56 per CPE; capacity 256
    lan_subnet:
      type: ipv4
      cidr: "10.0.0.0/16"               # capacity 65535

Then any leaf references the pool by name:

parameters:
  - path: Device.IP.Interface.2.IPv4Address.1.IPAddress
    value: "{wan_ipv4}"                 # 203.0.113.1, .2, ... per CPE
  - path: Device.IP.Interface.2.IPv6Address.1.IPAddress
    value: "{wan_ipv6}"                 # 2001:db8:1::1, ::2, ... per CPE
  - path: Device.IP.Interface.2.IPv6Prefix.1.Prefix
    value: "{delegated_prefix}"         # 2001:db8:cafe:100::/56, :200::/56, ...

Pool types:

• ipv4: Nth host in the CIDR. Capacity = 2^(32-prefixLen) - 1 (skips network base).
• ipv6: Nth host in the IPv6 CIDR. Same capacity formula on 128 bits.
• ipv6prefix: Nth /sublen prefix from super. Capacity = 2^(sublen - super.prefixLen). Models DHCPv6-PD-style ISP delegation.

Capacity is checked at LoadProfile¶

If fleet.count: 1001 references a pool that holds 1000, profile load rejects with:

fleet.count=1001 but pool "wan_ipv4" can't satisfy that many CPEs:
instance 1001 exceeds capacity 254 for cidr 203.0.113.0/24

Fail-fast at startup beats per-CPE failure mid-bootstrap.

Worked example: one container, 100 CPEs¶

deviceIdPaths:
  manufacturer: Device.DeviceInfo.Manufacturer
  oui:          Device.DeviceInfo.ManufacturerOUI
  productClass: Device.DeviceInfo.ProductClass
  serialNumber: Device.DeviceInfo.SerialNumber

fleet:
  count: 100
  serialPattern: "TEST-{i:04}"
  pools:
    wan_ipv4:
      type: ipv4
      cidr: "203.0.113.0/24"

parameters:
  - path: Device.DeviceInfo.Manufacturer
    value: "ACME"
  - path: Device.DeviceInfo.ManufacturerOUI
    value: "001122"
  - path: Device.DeviceInfo.ProductClass
    value: "HomeGateway"
  - path: Device.DeviceInfo.SerialNumber
    value: "TEST"

  - path: Device.IP.Interface.1.IPv4Address
    value: "{wan_ipv4}"                 # 203.0.113.1 .. 203.0.113.100
    writable: true
  - path: Device.Ethernet.Interface.1.MACAddress
    value: "AA:BB:CC:{cpe:mac:3}"        # AA:BB:CC:00:00:01 .. AA:BB:CC:00:00:64
    writable: true
  - path: Device.Hosts.Host.1.HostName
    value: "host-{cpe_id}"               # host-cpe-1 .. host-cpe-100
    writable: true

bin/cpe-sim --profile=this.yaml --acs-url=http://acs/cwmp

100 CPEs bootstrap in parallel. Each shows up at the ACS with a unique serial / MAC / IP / hostname. Logs carry cpe_id per line so grepping one CPE out of the noise is trivial.

CR routing per CPE¶

When fleet.count > 1 and --cr-bind-addr is set, the connection-request listener routes incoming requests by URL path: <cr-path>/<cpe-id> (e.g. /cr/cpe-3). Each CPE writes its full URL into the leaf named by --cr-publish-path so the next Inform reports the right ConnectionRequestURL.

When fleet.count == 1, the path is --cr-path verbatim. Single-CPE deployments unchanged.

Determinism¶

Per-CPE jitter and generator state derive from a per-CPE *rand.Rand seeded by FNV-64a hash of (rootSeed, cpeID). The root seed comes from --seed (or CPE_SIM_SEED); when 0, it's derived from time.Now().UnixNano() and logged at startup as root_seed=<N>. Pass that value back via --seed=<N> next run and every CPE's stream replays byte-for-byte.

Reserved pool names¶

Pool names must match [A-Za-z_][A-Za-z0-9_]* and cannot collide with built-in placeholders (cpe, cpe_id, i, base). Use snake_case for clarity.