v0.4.1 · Apache-2.0 · Go 1.26

A network load target
for the monitoring tools
you're writing.

l8opensim simulates up to 30,000 network devices, GPU servers, storage systems and Linux hosts on a single Linux host — each with its own IP, SNMP listener, SSH server, HTTPS REST endpoint and flow exporter. Built on TUN interfaces and network namespaces.

quick start →github ↗

noc@laptop — ~ — 132×36

noc@laptop:~$

devices / host

30,000max

device types

28in 8 cat.

mem / device

~1KB

parallel workers

500max

quick start

build from source · or pull with docker

01 · clonegit

$git clone https://github.com/labmonkeys-space/l8opensim.git

$cd l8opensim

02 · buildmake · go 1.26+

$make tidy

$make build

03 · runneeds root

$sudo ./go/simulator/simulator -auto-start-ip 10.0.0.1 -auto-count 100

or with docker

01 · pullno toolchain

$docker pull ghcr.io/labmonkeys-space/l8opensim:latest

02 · runneeds --cap-add=net_admin

docker run --rm -it \
  --cap-add=NET_ADMIN \
  --device=/dev/net/tun \
  --network=host \
  ghcr.io/labmonkeys-space/l8opensim:latest \
  -auto-start-ip 10.0.0.1 -auto-count 100

what's in the box

six pillars

30,000 devices

Tested scale on a single host. Parallel TUN pre-allocation, lock-free sync.Map for O(1) OID lookups, pre-computed next-OID mappings.

Protocols

SNMP v2c/v3 (MD5/SHA1 · DES/AES128), SSH with VT100, HTTPS REST, NetFlow v5 / v9 / IPFIX. sFlow v5 (experimental).

28 device types

Routers, switches, firewalls, servers, GPU servers (DGX/HGX), storage systems, Linux servers — across 8 categories.

GPU simulation

NVIDIA DGX-A100 / H100 / HGX-H200 with per-GPU DCGM OIDs — utilization, VRAM, temp, power, fan, SM/memory clocks.

Namespace isolation

Each device runs in the dedicated opensim network namespace with its own TUN interface and IP.

Dynamic metrics

100-point pre-generated sine-wave cycling for CPU, memory, temperature — correlated across related metrics.

device catalog

28 types · 8 categories · 341 resource files

core routers5

›Cisco ASR9K · 48
›Cisco CRS-X · 144
›Huawei NE8000 · 96
›Nokia 7750 SR-12 · 72
›Juniper MX960 · 96

edge routers3

›Juniper MX240 · 24
›NEC IX3315 · 48
›Cisco IOS · 4

dc switches2

›Cisco Nexus 9500 · 48
›Arista 7280R3 · 32

campus switches3

›Cisco Catalyst 9500 · 48
›Extreme VSP4450 · 48
›D-Link DGS-3630 · 52

firewalls4

›Palo Alto PA-3220 · 12
›Fortinet FortiGate-600E · 20
›SonicWall NSa 6700 · 16
›Check Point 15600 · 24

servers4

›Dell PowerEdge R750
›HPE ProLiant DL380
›IBM Power S922
›Linux Server · Ubuntu 24.04

gpu servers3

›NVIDIA DGX-A100 · 8×80GB
›NVIDIA DGX-H100 · 8×80GB
›NVIDIA HGX-H200 · 8×141GB

storage4

›AWS S3
›Pure Storage FlashArray
›NetApp ONTAP
›Dell EMC Unity

status & scale

what works · how big

stable6

■SNMP v2c/v3
■SSH (VT100)
■HTTPS REST (storage)
■NetFlow v5/v9/IPFIX
■TUN + netns isolation
■Web UI + REST API

experimental2

■sFlow v5 (synthetic)
■Layer 8 overlay

tested scale3

■30,000 concurrent devices / host
■~50 MB base + ~1 KB / device
■CPU: minimal in steady state

concurrent devices

30,000tested

device types

resource files

341json

world cities

98sysLocation

ssh commands

36+linux

documentation map

jump in

getting started

Build, bring up a small fleet, run in Docker.

operations

Scale to 30k, tune the opensim namespace, flow export and SNMP traps.

reference

Architecture, CLI flags, REST API, device-type tables, protocol details.

gpu simulation

DGX/HGX simulation, DCGM OID layout, pollaris parser.

→ get started

Spin up tens of thousands of devices in seconds.

Apache-2.0. No agents, no cloud, no per-device fees. Just TUN interfaces and a little Go.