Interconnect, served LANs and subnet routers

Join two networks. Merge neither.

Three patterns let a machine reach hosts it could not reach before, and all three are often described as “connecting networks”. They answer different questions. Atlas Interconnect joins two separate Atlas networks, each with its own keys, members and operator, at one host called the seam, and lets an allow-listed set of hosts on each side reach the other. A served LAN brings a plain LAN, whose hosts run nothing, into one Atlas network through a node with an interface on it. A subnet router does the same for a flat overlay run by one coordination service. This page explains each mechanism, the two modes Interconnect offers, how the three patterns differ and which one fits which case.

Opt-inMapped mode · Prefix mode · Served LANs
01 · What Interconnect is

Two networks meet at one host.

An Atlas network is sealed: its own keys, its own address space, forwarding only among its own members. To join two of them, an operator runs one daemon per network on a single host and connects the two daemons with a link that exists only inside that host. Everything that crosses is data between hosts both sides have agreed to expose. Nothing about either network’s structure crosses at all.

NETWORK A · 10.0.101.0/24 SEAM HOST NETWORK B · 10.0.9.0/24 Phone10.0.101.7 · mobile app Ground station10.0.101.2 · exported Radio relay10.0.101.5 daemon Amember of A daemon Bmember of B Sensor10.0.9.14 Partner drone10.0.9.2 · exported Operator laptop10.0.9.30 ADVERTS · GOSSIP · KEYS stop here, in both directions local link · loopback only · data, no routing EXPORT A gates both ways EXPORT B gates both ways 1 · ONE DAEMON PER NETWORK each admitted and revoked by its own operator 2 · EXPORT GATES BOTH WAYS only listed hosts send or receive across 3 · NOTHING CHAINS a packet that crossed once never crosses again members of either network install nothing and configure nothing
THE SEAM HOSTruns one daemon in network A and one in network B, joined by a loopback-only link
WHAT STOPS THEREadverts, gossip and keys: neither network learns the other’s structure
WHAT CROSSESpackets between hosts on each side’s export list, never onward to another seam
The seam host

One daemon per network, side by side

Each Atlas daemon owns one tunnel interface, one identity and one address space. The seam host runs two of them: daemon A is an ordinary member of network A, admitted under A’s rules, and daemon B is an ordinary member of network B under B’s. Neither network grants the other anything. Either operator ends the join by revoking the seam host’s membership, exactly as it would revoke any other member.

The local link

A link that only exists inside one host

The two daemons peer over a link of type local, which config load refuses to bind anywhere but loopback. It is an ordinary Atlas link with its usual encryption, so the seam reuses the transport rather than adding a second one; it simply cannot be reached from outside the host.

Export, both ways

Each side lists what takes part

Each daemon carries an export list naming the hosts or prefixes of its own network, always inside its own subnet, that may take part. The list gates both directions: an unexported host can neither be reached from the other side nor send across to it. Each operator writes only its own list, so neither side can widen what the other exposes.

No chaining

Parallel seams, never transit

Nothing is redistributed, and a packet that arrived across one seam is never sent across another, so seams cannot be strung together into a path nobody approved and loops cannot form. Two seam hosts joining the same pair are two parallel seams: every member agrees on one owner of each address the seams serve, and the other takes over within seconds when the first one becomes unreachable.

Members untouched

Two control planes that never meet

Link-state adverts, full topology exchanges and membership gossip are never sent across the seam and are ignored if they arrive; the seam is not even listed as a link. Neither network gains a node of the other in its topology, learns the other’s keys or sees its admission decisions. Members keep the configuration they have.

02 · The two modes

Map a host, or route a prefix.

A seam carries traffic in one of two ways, chosen per entry. Mapped mode makes a host of the other network appear at an address inside your own subnet, so members reach it with what they already have. Prefix mode routes a whole prefix of the other Atlas network through the seam with real addresses and nothing rewritten, so its members are reached at the addresses they already have. One seam can carry both at once.

Mapped mode: the far host appears inside your own subnet

Each host the operator maps is given an address inside this network’s own /24, chosen by this network’s operator. The seam daemon advertises that address as one it serves, so every member reaches it with the subnet route and the shortest-path tree it already has. At the seam the destination is rewritten to the host’s real address on the other side, the source is rewritten to the sender’s mapped address there, and IPv4 and TCP/UDP checksums are repaired in flight. Each side writes only its own map, so nobody renumbers anybody.

NETWORK A · 10.0.101.0/24 SEAM HOST NETWORK B · 10.0.9.0/24 Phone10.0.101.7 · fixed /24 daemon Aserves 10.0.101.202 daemon Bserves 10.0.9.107 Ground station10.0.9.2 · unchanged MAP IN A .202 ⇄ 10.0.9.2 MAP IN B .107 ⇄ 10.0.101.7 1 · THE PHONE SENDS 10.0.101.7 → 10.0.101.202 inside its own subnet: no route to add 2 · THE SEAM REWRITES to 10.0.9.2, from 10.0.9.107 checksums repaired, exports checked 3 · B DELIVERS 10.0.9.107 → 10.0.9.2 an ordinary host in B’s own subnet no route, flag or update on any member · older builds and the mobile app included
THE PHONEsends to 10.0.101.202, an address inside its own fixed /24
THE SEAMrewrites the destination to 10.0.9.2 and the source to 10.0.9.107
NETWORK Bsees an ordinary host in its own subnet; no member changed

Members need nothing

The mapped address is one more address in the member’s own subnet. Phones, whose app has a fixed address space, reach the far side as easily as a laptop does, and so do members running older builds.

Your numbering stays yours

Nothing of the far network’s numbering enters your routing: each host you map takes an address you choose inside your own subnet. The two networks joined at a seam must not share an address space; partners that happen to be numbered alike as each other each join you through their own seam.

Networks run by someone else

The other side does not change its plan for you. It admits the seam host as a member and writes its own export list and map; you write yours. Neither side needs the other’s keys or topology.

Prefix mode: the other network’s members, at their real addresses

Some joins need more than a handful of hosts: a partner’s ground-control network whose ROS 2 bridge node, video node and operator consoles must all be reachable, or a second organisation’s whole network. Prefix mode routes a prefix of the other Atlas network, its members’ tunnel addresses, through the seam with real addresses in both directions and nothing rewritten. The seam daemon carries the prefix as an imported route in its network’s signed link-state adverts, and each member’s daemon installs it on its own tunnel interface ([routing] install_learned_routes, on by default), so the member reaches a far host by the address that host already has. The reply path is the mirror image: the far seam daemon imports this network’s prefix with its own route entry and advertises it in its network. A prefix crosses only when both sides list it, one side’s route and the other side’s export, for each direction.

NETWORK A · 10.0.101.0/24 SEAM HOST NETWORK B · 10.0.9.0/24 signed advert: 10.0.9.0/25 signed advert: 10.0.101.0/24 Operator laptop10.0.101.4 · route on atlas0 daemon Aimports 10.0.9.0/25 daemon Bimports 10.0.101.0/24 ROS 2 bridge10.0.9.10 · Atlas member Video node10.0.9.20 · Atlas member 1 · BOTH SIDES LIST IT A routes 10.0.9.0/25, B exports it and the mirror image for the replies 2 · ADVERTISED, INSTALLED each member’s daemon adds the route on its own tunnel interface, nowhere else 3 · DELIVERED real addresses, both ways nothing rewritten at the seam two seam hosts on one prefix fail over in seconds · routes leave members about two minutes after the last seam host goes
BOTH SIDES LIST ITA routes B’s 10.0.9.0/25 and B exports it; B routes A’s prefix for the replies
ADVERTISED, INSTALLEDeach seam daemon advertises the import; members put the route on their own tunnel interface
DELIVEREDreal addresses both ways, nothing rewritten; far members see the caller’s own address

Real addresses

Far members are reached at the addresses they already have, and they see the real address of the member that called them. Applications that write addresses into their own payloads, such as stream URLs and configured peer lists, keep working without translation.

Learned, not configured

The prefix arrives in the same signed adverts that carry the rest of the network’s routing, and each daemon installs it on its own tunnel interface. Withdraw it at the seam and it leaves every member. A member that already has a route overlapping the prefix keeps its own and logs the skip.

Members, not LANs

What prefix mode routes is the other network’s own members. A LAN of hosts that do not run Atlas, behind a node of your own network, is a served LAN, and a served LAN never crosses a seam. The mobile app’s address space is fixed, so phones reach the other network through mapped mode.

Both modes on one seam. A partner’s ground-control network can be routed to your laptops by prefix, while your phones, whose address space is fixed, reach its operator console through a mapped address. Each entry chooses its mode; the export lists, the isolation of the two control planes and the no-chaining rule apply to every entry alike. Two seam hosts serving the same addresses fail over within seconds in either mode.
net-a.toml — one seam, both modes
[peer.interconnect]
export = ["10.0.101.0/24"]         # this network's hosts that take part

[[peer.interconnect.map]]          # mapped mode
remote = "10.0.9.130/32"
local  = "10.0.101.130/32"

[[peer.interconnect.route]]        # prefix mode: the other network's prefix, real addresses
prefix = "10.0.9.0/25"

# on every member: the default
[routing]
install_learned_routes = true
03 · Served LAN prefixes

A plain LAN, served into one network.

Printers, cameras, a file server, a controller: the hosts on a site’s flat network run nothing, and nothing is installed on them. Inside one Atlas network, a node with an interface on that LAN serves its prefix into the mesh. Every member learns the route from the serving node’s signed advert and installs it on its own tunnel interface, so a laptop behind a radio relay reaches the camera at the camera’s own address. This is the role a subnet router plays, and a different thing from a seam: a served LAN extends one network to hosts that run nothing, where a seam joins two networks of members.

ATLAS MESH · 10.0.101.0/24 SERVING NODE PLAIN LAN · 192.168.10.0/24 signed advert: 192.168.10.0/24 Operator laptop10.0.101.4 · learns the route Radio relay10.0.101.5 · forwards Drone10.0.101.9 · learns the route Ground station10.0.101.2 · serves on eth1eth1 192.168.10.1 LISTED SOURCES ONLY the rest is refused and counted Printer192.168.10.20 · runs nothing Camera192.168.10.31 · runs nothing LAN gatewayroute 10.0.101.0/24 via 192.168.10.1 the one static route, for real addresses only 1 · SERVED one [[serve]] block on one node nothing installed on the LAN 2 · LEARNED every member installs the route itself nearest serving node · failover in seconds 3 · BOTH WAYS masquerade on: the LAN sees the node off, plus the gateway route: real addresses a served LAN never crosses an Interconnect seam · members configure nothing
SERVEDone node with an interface on the LAN serves its prefix; nothing is installed on the LAN
LEARNEDevery member installs the route from the signed advert and uses the nearest serving node
BOTH WAYSthe mesh reaches the LAN; with one static route on the LAN gateway, the LAN reaches the mesh at real addresses
One node

One block, on the serving node only

A [[serve]] block names the LAN’s prefix, outside the mesh subnet, the serving node’s interface on that LAN, whether the node masquerades, and who may cross. Nothing else in the network is configured: members keep the default, and the LAN’s hosts are not touched. In a fleet run from signed records, a node’s record bounds which prefixes it may serve, so no member can pull a LAN’s traffic to itself by announcing it.

Learned routes

Every member installs the route itself

The serving node floods the prefix in its signed advert, and each member’s daemon installs it on its own tunnel interface ([routing] install_learned_routes, on by default). A member reaches every host on the LAN by the address that host already has, through as many mesh hops and bonded links as the path takes. atlasd status and the dashboard’s learned-routes table name each route a member holds as a served LAN or an Interconnect import.

Masquerade

Masquerade, or one static route

With masquerade on, the default, the serving node translates the source of what it forwards onto the LAN to its own LAN address: the LAN’s hosts answer the node and need no route back. With masquerade = false nothing is translated in either direction; one static route on the LAN’s gateway, the mesh subnet via the serving node’s LAN address, brings the replies back and lets the LAN’s hosts open connections to the members allowed to cross. Both ends then see each other’s real address.

Who may cross

Listed sources only; the rest is refused and counted

for names who may reach the LAN: the whole mesh by default, or named records, node ids or tunnel addresses. The serving node forwards only from those sources to the LAN and only from the LAN back to them, and drops everything else; the refusals are counted in its status, its snapshot and its journal.

Two servers

The nearest one, and the other within seconds

Two nodes may serve the same LAN. Each member routes to the nearest one it can reach, and traffic moves to the other within seconds when one goes away; a clean stop withdraws the route from the members before the serving node removes its forwarding rules.

The seam

A served LAN never crosses an Interconnect seam

A served prefix lies outside the network’s own subnet, and a seam exports only addresses inside it, so a served LAN cannot be exported to a partner network; a packet that arrived through a seam is dropped at the serving node as well. This is not a setting. A served LAN belongs to the network its node is a member of.

The serving node runs Linux and forwards through a firewall table of its own; the tunnel carries IPv4; each LAN interface carries one served prefix.

ground-station.toml — the serving node
[[serve]]
prefix = "192.168.10.0/24"
iface = "eth1"
masquerade = true          # false: real addresses both ways
for = ["*"]                # the whole mesh, or records / tunnel addresses

# on the LAN gateway, only with masquerade = false:
# ip route add 10.0.101.0/24 via 192.168.10.1
04 · A subnet router

One flat overlay, one coordination service.

The most familiar way to reach a plain LAN from far away is a flat overlay: one private network built from devices that run its client, with a coordination service that hands out keys, addresses and access policy while the traffic flows between the devices themselves. A subnet router is one of those devices with an interface on a LAN. It advertises the LAN’s prefix, and the overlay’s devices reach the LAN through it. What follows describes the pattern, not any one product; deployments differ in the details.

Relayfallback · packets stay sealed Coordination servicekeys · addresses · policy to every device · carries no traffic ONE OVERLAY · ONE POLICY Laptopruns the client Serverruns the client Phoneruns the client Subnet routeradvertises 192.168.1.0/24 LAN · 192.168.1.0/24 Camera Controller run nothing · reached via the router 1 · ADVERTISE the router announces 192.168.1.0/24 a setting on the router itself 2 · APPROVE an administrator approves the route in the overlay’s one policy 3 · ACCEPT the overlay’s devices install it some clients opt in first
CONTROLa coordination service hands out keys, addresses and policy; it carries no traffic
DATAencrypted tunnels between devices: direct where NAT allows, else through a relay
SUBNET ROUTERadvertises a LAN prefix; an administrator approves it; the devices install the route
Control plane

A coordination service for keys and policy

Devices register with the coordination service, which shares each device’s public key and address with the others and pushes the access policy to every device. Private keys stay on the devices and the policy is enforced on each of them, so the service distributes configuration rather than carrying traffic.

Data plane

Direct tunnels, with relays as fallback

Devices connect to each other over encrypted tunnels, traversing NAT to reach a direct connection where they can. Where a direct connection fails, typically behind blocked UDP or hard NAT, the traffic goes through a relay that forwards packets which stay encrypted end to end. Each connection takes one path at a time over the IP connectivity the device already has.

Subnet routers

A LAN prefix, advertised, approved and accepted

The router advertises one or more prefixes. In most designs an administrator approves each one, or a policy rule approves it automatically, and the overlay’s devices install the approved routes; some clients opt in before they use them. The LAN’s hosts need no software. By default the router translates the source address, so the LAN’s hosts see traffic coming from the router; keeping the real source needs a return route for the overlay’s range on the LAN. Several routers can usually advertise the same prefix for high availability.

One trust domain

One overlay, one set of administrators

Everything a subnet router brings in joins the one overlay: its coordination service, its policy and its administrators. Working with another organisation usually means bringing that organisation’s users or devices into the same overlay, under the same policy, rather than joining two overlays that each keep their own.

05 · How the three differ

One network extended, or two networks joined.

Most rows below follow from one design choice. A served LAN and a subnet router make one trust domain reach further, to hosts that run nothing. Interconnect keeps two trust domains whole and lets a narrow, allow-listed flow pass between them at a host both have admitted.

Interconnect seamServed LANSubnet router on a flat overlay
The other sideMembers of another Atlas network, run by its own operatorA plain LAN whose hosts run nothing, behind a node of your own networkA plain LAN whose hosts run nothing, behind a device of the overlay
Who configuresThe seam host only: each operator writes its own export list, map entries and routes for its own daemon. Members configure nothingThe serving node only: one [[serve]] block. Members keep the default; the LAN needs nothing, or one static route on its gatewayThe router advertises the prefix, an administrator of the overlay approves it in most designs, and some clients opt in to accept it
AddressesMapped mode: a far host appears at an address you choose inside your own subnet, and the sender appears at its mapped address on the far side. Prefix mode: real addresses both ways, nothing rewrittenMembers reach the LAN’s hosts at their real addresses. With masquerade on, the default, the LAN sees the serving node; off, with the gateway route, real addresses both waysDevices reach the LAN’s hosts at their real addresses. By default the router translates the source, so the LAN sees the router; the real source needs a return route on the LAN
DirectionBoth ways, for the hosts on each side’s export list; in prefix mode each direction crosses only when one side’s route and the other side’s export both list itMesh to LAN, for the sources the for list names. LAN to mesh as well with masquerade off and the gateway routeOverlay to LAN. LAN to overlay where source translation is off and the return route is in place
Route originThe seam daemon. Mapped mode: it serves each mapped address inside the subnet members already route. Prefix mode: it imports the far prefix and advertises it in its network’s signed link-state advertsThe serving node floods the prefix in its signed advert, and every member’s daemon installs it on its own tunnel interface; in a record fleet the node’s record bounds what it may serveThe coordination service distributes the approved route to the overlay’s devices
Trust domainsTwo networks, each with its own keys, admission and operator; the seam host is a member of both, and either side ends the join by revoking itOne network; the serving node is one of its membersOne overlay: one coordination service, one policy, one set of administrators
FailoverTwo seam hosts serving the same addresses: the other takes over within seconds, in both modesTwo nodes serving the same LAN: each member uses the nearest, and traffic moves within seconds when one goes awaySeveral routers can usually advertise the same prefix
PathAlways through the seam host; on each side, that network’s own routed paths and bonded linksThe mesh’s own routed paths, across as many hops and bonded links as the topology takesDirect between devices where NAT allows, else through a relay; LAN traffic goes through the router

The subnet-router column describes the common pattern of a flat overlay with a coordination service; a given product can differ in the details. Where a specific deployment differs, bring it to the conversation.

Nexus Atlas: Beyond the Mesh4:08 · plays from YouTube
06 · When to pick which

Pick by who runs the other side.

The deciding question is rarely the tunnel. It is who runs the hosts on the far side, and whether they run anything at all. Each card is written for the same reader: someone choosing how to let one group of machines reach another.

An Interconnect seam

Pick it when
  • The far side is a network someone else runs, and neither side will merge into the other or share a control plane.
  • Members cannot be touched: phones with a fixed address space, relays on masts and members on older builds reach the far side through mapped mode with nothing changed.
  • A whole prefix of the other network must be reached at real addresses, learned from the signed adverts rather than configured per device.
  • Exposure has to be allow-listed on both sides and must never chain through to a third network.
What it asks
  • Every packet between the networks passes through a seam host, and the seam host is trusted by both sides: end-to-end sealed payloads are opened there and re-sealed.
  • Mapped mode needs one map entry per far host, and a host has a different address on each side.
  • The two networks must not share an address space, in either mode.
  • A seam carries members of the two networks only.

A served LAN

Pick it when
  • The hosts on a site’s flat network cannot run anything, such as printers, cameras, controllers or a file server, and the site belongs to your own network.
  • Members should reach those hosts at their own addresses over the mesh’s routed, bonded paths, with the route learned rather than written on every member.
  • Only named sources may cross, and every refusal should be counted.
  • A second node on the same LAN should take over within seconds.
What it asks
  • A Linux node with an interface on the LAN.
  • With masquerade on, the LAN sees the serving node rather than the caller; real addresses need one static route on the LAN’s gateway.
  • One served prefix per LAN interface, and the tunnel carries IPv4.
  • A served LAN stays inside its own network and never crosses a seam.

A subnet router on a flat overlay

Pick it when
  • Laptops, servers and cloud hosts of one organisation connect over ordinary internet, directly device to device where the network allows.
  • Sign-in through an existing identity provider and a coordination service someone else runs are what the fleet needs most.
  • A LAN that runs nothing must be reached by that same fleet, under the same policy.
What it asks
  • One trust domain: everything joins one overlay under one set of administrators.
  • In most designs each route is approved, and some clients opt in before they use it.
  • By default the LAN sees the router’s address, not the caller’s.
  • Each connection rides one path at a time over the IP connectivity the device already has.
07 · Why two modes

Where two modes beat one subnet router.

A subnet router answers one question well: how the devices of one network reach a LAN that cannot run anything. Inside one Atlas network a served LAN answers it the same way; a seam does not, by design. The cases Interconnect is built for ask different questions, and the two modes together answer them without asking any member, or any partner, to change.

Fleets with phones and radios

A team’s phones, a relay on a mast and a drone on a serial radio cannot all be asked to accept new routes before a mission. Mapped mode makes the partner’s console one more address in their own subnet; prefix mode brings a partner’s whole ground-control network to the laptops whose daemons learn it from the adverts. Neither waits on a route approval or a client setting, and the traffic keeps riding the bonded links it already uses.

Two organisations, no shared control plane

An agency and a contractor rarely agree to one set of administrators over both their devices. With a seam, each keeps its own network, keys and admission. Each admits one host and writes one export list, and either side ends the join by revoking a single member.

Numbering you do not control

Partners that use the same ranges as each other are routine in coalition and multi-vendor work. Mapped mode places each far host at an address chosen inside your own subnet, so nothing of a partner’s numbering enters your routing, and partners numbered alike as each other each join you through their own seam. The one rule: the two networks joined at a seam must not share an address space.

Allow-listed exposure

A partner should see the ground station and the video encoder, not the fleet. Export lists on both sides name exactly what takes part, gate both directions, and stop at the seam: nothing is redistributed, nothing chains, and neither network learns the other’s shape.

Where a flat overlay is the better answer. Laptops, servers and cloud hosts of one organisation over ordinary internet, with direct device-to-device connections, sign-in through an existing identity provider and a coordination service someone else runs: that is the problem a flat overlay solves, and it solves it well. Some overlays also reach sites that use identical ranges, which a seam does not join. A LAN of devices that cannot run any software is reached either way: by a subnet router on an overlay, and by a served LAN inside one Atlas network. Interconnect is for the case where the networks belong to different owners, the members cannot be touched, and the links underneath are radios as often as fibre. The wider comparison is on How Atlas compares.
Request an evaluation

Two networks that must talk, and neither will merge?

Tell us who runs each side, how each is numbered and what has to cross. The answer is a seam placement, a mode per flow and an export list for each side.