Networking & Fleet Management

DHCP vs. static addressing

Every ASIC has one Ethernet port and no screen, so the network is the only way in. Out of the box, Bitmain, MicroBT, and Canaan control boards all default to DHCP: on boot they ask the site DHCP server for an address and accept whatever lease they're handed. That gets a machine online fast, but the address can change on the next lease renewal, which breaks any tool that expects a machine to stay put. At hyperscale you pin addresses one of two ways: a static IP configured on the miner itself, or a DHCP reservation that ties a MAC address to a fixed IP on the server. Reservations are usually the better call for a big fleet because you manage all addressing centrally instead of walking to each machine.

Subnets and VLANs for a mining LAN

A mining floor is thousands of chatty embedded-Linux devices on cheap hardware. You don't want them on the same flat network as office PCs or management servers. Segment the fleet onto its own VLAN and subnet: it contains the broadcast traffic ASICs generate, keeps a compromised or misbehaving miner away from business systems, and gives you clean address blocks to reason about. Plan the address space before you rack anything — a common pattern is a subnet (or VLAN) per row, PDU group, or container, so an IP tells you roughly where the machine physically is. That mapping is gold when you're chasing a fault.

• Size subnets for growth — a /24 holds ~254 hosts, so a dense row or container often needs a /23 or larger, or one VLAN per row.
• Use managed switches once you pass ~20 machines: you get VLAN tagging, SNMP monitoring, port mirroring for diagnostics, and uplink redundancy.
• Run real Ethernet to every miner. No Wi-Fi bridges, no powerline adapters — they drop packets and hide faults.
• Keep miner, management, and PDU/networking-gear traffic on separate VLANs so you can still reach the infrastructure when the miner VLAN is flooded.

Finding a miner on the network: IP-reporter tools

When a machine is on DHCP and nothing shows its address, every maker gives you a broadcast-based way to find it. You press a physical button on the control board and the miner blasts a packet containing its hostname, MAC, and current IP; a listener tool on a PC in the same broadcast domain catches it and shows the address. The button and tool names differ by maker, but the mechanism is identical. This only works when your PC is on the same VLAN/subnet as the miner — the broadcast won't cross a router.

The cgminer/bmminer API on TCP 4028

Nearly every stock ASIC firmware descends from cgminer, and it exposes a JSON control/monitor API on TCP port 4028 (bmminer on Antminers, btminer on Whatsminers, and cgminer-derived stacks on Avalon and custom firmware all speak it). This is how monitoring platforms poll a fleet without logging into web UIs: they open 4028, send a command like {"command":"summary"}, and parse the JSON reply. You can test any machine by hand with netcat: echo '{"command":"stats"}' | nc <miner-ip> 4028. Read commands are open by default; privileged/write commands (switch pools, restart) require the firmware to be started with API access allowed for writing.

Pools, workers, and DNS — where hashrate quietly disappears

A miner points at one or more pools via a stratum URL (for example stratum+tcp://host:port) plus a username and a worker name. The username is usually the account or wallet; the worker name identifies that specific machine on the pool dashboard so you can tell rig-A12 apart from rig-A13. Most operators configure three pools per miner — a primary and two backups — so the machine fails over automatically if a pool goes unreachable. Two things silently cost you money here: a bad worker/username, and DNS that can't resolve the pool hostname.

• Wrong or blank worker name: the pool may still show hashrate but no worker (or 'undefined'), and that hashrate isn't cleanly attributed to you. Worst case, a typo in the username sends the reward to the wrong account entirely.
• Keep worker names unique and machine-mapped (many pools allow 1–15 alphanumeric characters, no special symbols) so a dead worker on the dashboard points you straight to the physical unit.
• DNS matters: the miner needs a working gateway and DNS server to turn the pool hostname into an IP. Link up + valid IP but a blank/unreachable DNS = the stratum host never resolves and the machine mines nothing.
• A stratum port blocked outbound at the firewall looks the same from the floor — link is fine, pool is 'Dead'. Confirm the pool port is open egress from the miner VLAN.

Fleet management platforms

Past a few dozen machines you stop touching miners individually and run a fleet platform. They all do the same core job — poll thousands of miners (mostly over the 4028 API), surface hashrate/temps/errors/offline status, alert on faults, and push bulk actions like reboot, pool change, firmware update, and power/curtailment control. They differ mainly in whether they need an on-site agent and how deep the control goes.

How fleet tools discover miners

A cloud dashboard can't reach machines sitting on a private mining VLAN, so most platforms either run a local agent (Foreman's Pickaxe, minerstat/Hive's ASIC Hub) that lives on the LAN and relays data out, or connect directly to 4028 when they're on the same network (Awesome Miner). Discovery itself happens a few ways, and big sites use more than one.

• Broadcast capture ('interactive add'): the tool listens while you press miners' IP-report buttons and grabs each broadcast — good for onboarding a rack.
• Active IP-range scan: the tool sweeps a subnet and flags anything answering on 4028 (or the web/SSH port). Anything not responding on 4028 simply won't be found.
• ARP / MAC resolution: matching IPs to MAC addresses to catch machines and spot duplicates.
• Import from infrastructure: pull a site map from managed switches (which port = which location) or from PDUs, so the dashboard knows physical position, not just IP.

Common connectivity faults and triage

Most network calls on the floor fall into a handful of patterns. Read the symptom, not the panic — the single biggest tell is scope: one machine is a machine problem; a whole row dropping at once is almost never the miners.