Frontierthe named edge

Carrier & Inter-Networking

Carrier inter-networking is Nikita Tarikin's one named frontier: BGP path-vector mechanics, MPLS/VPLS, and traffic engineering are studied at concept level toward the MikroTik Certified Inter-Networking Engineer certification (MTCINE, not yet taken) but not deployed in production — the honest edge of the map.

The boundary of the map: concept-aware, deployment-blind. Below, the leaf ledger marks exactly where the knowledge stops and the unbuilt begins.

1.9
Knowledge K̄: 1.3
Deployment Ē: 27
leaves mapped: 17
know ≠ run splits

The map

Every carrier-grade leaf in the official MikroTik inter-networking curriculum, placed by what is known against what is run in production. Here the marks gather in the bottom-left — the low-knowledge, low-deployment corner — and the mastery corner stays empty. That emptiness is the honest shape of a frontier: grasped at the headline, not yet built.

Read left-to-right as knowledge, bottom-to-top as deployment. The accent-orange marks and their short drop-lines are this frontier’s daylight — concepts held that have never been deployed. Unlike every other domain on this site, there is no dense top-right cluster; that empty mastery corner is the shape of a frontier.

The topography

The full leaf-level ledger, grouped by topic — BGP, MPLS, and traffic engineering. Each row carries its own two verdicts and the field note that earned them, quoted verbatim from the assessment — including the ones that simply read “no idea” or “black box.”

Leaf	Knowledge	Deployment	Field note
BGP		K̄ 2.5 · Ē 1.7 · 11 leaves · 9 splits
Autonomous Systemsstudied ahead	3Confident	2Aware	AS = one administrative/routing-policy domain identified by an ASN. Knows AS_PATH is also BGP's loop prevention. (supervised ROS6).
BGP Mechanics & Path-Vector Algorithmstudied ahead	3Confident	2Aware	knows ROS6 instance/peer -> ROS7 template/connection + new /routing/filter/rule (RPKI-aware) -- aware of the rewrite that disassembled his ROS6 setup ~4yr ago (tracks RouterOS dev), even without rebuilding on it
Transport & Packet Typesstudied ahead	3Confident	2Aware	BGP = TCP/179, neighbors MANUALLY configured — no auto-discovery. The defining BGP trait.
iBGP / eBGP (Stub & Non-Stub)studied ahead	3Confident	2Aware	eBGP = inter-AS (prepend + re-advertise); iBGP = intra-AS (no AS-path change + no iBGP->iBGP re-advert -> full-mesh/RR). Residual from supervised ROS6 ~8yr ago
Multi-Hop & Loopbackstudied ahead	3Confident	2Aware	Peer to loopbacks so the session survives a link failure if any path exists. Implies the iBGP-over-loopback + IGP-underneath model.
Route Distribution & Filtersstudied ahead	3Confident	2Aware	Prefix advertise/accept control = routing filter rules (in/out chains) on prefix/AS_PATH/communities. Configured these in ROS6 BGP.
Best-Path Selectionstudied ahead	3Confident	2Aware	local-pref = outbound (internal to the AS); inbound TE = MED (ignorable) / AS-path-prepend / more-specifics, only a HINT (the neighbor's local-pref overrides) -- knows the real asymmetry
Prefix Attributesstudied ahead	3Confident	2Aware	AS_PATH as a core path attribute. BUT ceiling — missed the NEXT_HOP-unchanged-across-iBGP gotcha (chose iBGP-rewrites-to-loopback; real = next-hop stays the external peer → next-hop-self needed). Knows the attributes, not the deep iBGP-propagation subtlety.
Route Reflectors	1Learning	1Learning	Honest don't-know: 'no idea'. (RR re-advertises iBGP routes to clients → collapses the iBGP full-mesh to hub-spoke.) The carrier scale-policy edge — never built it. Matches communities.
Confederations	1Learning	1Learning	Honest don't-know: 'no idea'. (Confederations = split one AS into private sub-ASes to reduce the iBGP full-mesh — the other fix RRs also solve.) Carrier scale-policy edge.
Communitiesstudied ahead	2Aware	1Learning	EDGE: BGP communities 'no idea' (NO_EXPORT/policy-tags never used); route-reflector originator-id/cluster-list never touched. The carrier POLICY/scale depth beyond his basic supervised setup
MPLS		K̄ 1.5 · Ē 1.0 · 12 leaves · 6 splits
MPLS Basicsstudied ahead	2Aware	1Learning	MPLS forwards by a short label swapped hop-to-hop, not by IP-prefix lookup. The one MPLS concept he genuinely holds at headline level — 'I just heard MPLS delivery differs from classic routing'.
Static Label Mapping	1Learning	1Learning	BLIND. Batch-3 'correct' pick was by OBVIOUSNESS, disavowed: 'looks the most obvious; never touched MPLS in real life, the whole MPLS section is a black box for me'. Recorded is_correct=1 + don't-know=1 → Learning.
LDP (Label Distribution Protocol)	1Learning	1Learning	BLIND. The batch-3 'correct' pick was ELIMINATION (he flagged it: 'guessable strawmans'); the HARD re-probe (LDP downstream-unsolicited vs RSVP-TE/segment-routing/BGP-LU) returned 'no idea'.
Penultimate-Hop Popping	1Learning	1Learning	Honest don't-know: 'no idea'. (Penultimate-hop popping — the second-to-last LSR removes the label so egress does one IP lookup, not two.).
LDP-Based VPLSstudied ahead	2Aware	1Learning	conceptual headline only (LAN-over-MPLS); rides the blind MPLS -> necessarily thin, zero hands-on
Bridge Split-Horizonstudied ahead	2Aware	1Learning	Knows L2 bridge split-horizon cold (MTCSWE interface-horizon Teach) but NOT its VPLS application (PE split-horizon across pseudowires to prevent loops in the full-mesh of PWs). Adjacent-knowledge, MPLS-context blind.
VPLS Control Word	1Learning	1Learning	Deep VPLS internal (the control word prevents frame-ordering/ECMP misparse of pseudowire payloads). VPLS-concept is Aware but this internal is blind.
L2MTU & Fragmentationstudied ahead	2Aware	1Learning	Strong general MTU/L2MTU mastery (MTCSWE + MTCRE Teach) but NOT the MPLS-specific label-stack-overhead-on-MTU specifics (each label = 4 bytes; VPLS/L3VPN stacks). Adjacent-knowledge, MPLS-context blind.
BGP-Based VPLSstudied ahead	2Aware	1Learning	VPLS-concept GENUINE: picked 'multipoint L2 Ethernet bridge over MPLS' against HARD distractors (L3VPN-VPNv4 / VRF / RSVP-TE-LSP). Knows VPLS=L2 vs L3VPN=L3. The BGP-signaled-VPLS control-plane specifically not probed → concept Aware, mechanism Learning.
VRF & Route Leaking	1Learning	1Learning	Wrong — picked 'static inter-VRF route on the PE' (real = Route-Target import/export between VRFs). Knows L3VPN exists (VRF+RD) but not the RT-leaking machinery.
L3VPN / BGP L3 Tunnelsstudied ahead	2Aware	1Learning	L3VPN-concept GENUINE: picked 'per-customer VRFs + Route-Distinguisher makes overlapping prefixes unique in MP-BGP' against HARD distractors (RSVP-TE/VPLS/NAT). Holds the overlapping-address-isolation concept. The mechanism depth (RT, VPNv4 NLRI) is blind.
OSPF as CE-PE	1Learning	1Learning	Honest don't-know: 'thats overkill for me'. (OSPF as CE-PE = PE redistributes customer OSPF into MP-BGP VPNv4 across the core — the OSPF super-backbone/sham-link model.).
Traffic Engineering		K̄ 1.5 · Ē 1.0 · 4 leaves · 2 splits
Traffic Engineering Conceptsstudied ahead	2Aware	1Learning	Concept headline only: 'MPLS-TE can consider bandwidth/saturation; I guess it allows bandwidth reservations somehow; I'm blurry here'. Has the TE-steers-by-bandwidth idea, not the machinery.
RSVP (Static & Dynamic Path)	1Learning	1Learning	BLIND — the TE concepts probe showed he doesn't know RSVP-TE is the LSP-signaling + bandwidth-reservation protocol ('blurry').
CSPF (Constrained Shortest Path First)	1Learning	1Learning	Honest don't-know: 'no idea'. (CSPF = SPF that prunes links failing the constraint — bandwidth/affinity — before computing shortest path.).
Bandwidth Allocation & Limitationstudied ahead	2Aware	1Learning	From the TE-concepts probe he holds the 'TE reserves bandwidth' headline but not the allocation/limitation mechanics (per-LSP reservation, available-bandwidth accounting in the TE-DB).

The edges

This is where the map ends. On the stronger leaves — autonomous systems, eBGP versus iBGP, route filtering — there is a real conceptual grip, residue of a supervised RouterOS build years ago that the version-seven rewrite has since disassembled. On the harder ones — route reflectors, confederations, LDP, RSVP traffic engineering — the recorded verdict is the owner’s own words: “no idea,” or “the whole MPLS section is a black box for me.” None of it has been run at carrier scale.

Knowledge ahead of deployment

Known deeper than it currently runs — named, not rounded over.

Autonomous Systemsknow 3 Confident→run 2 Aware
AS = one administrative/routing-policy domain identified by an ASN. Knows AS_PATH is also BGP's loop prevention. (supervised ROS6).
BGP Mechanics & Path-Vector Algorithmknow 3 Confident→run 2 Aware
knows ROS6 instance/peer -> ROS7 template/connection + new /routing/filter/rule (RPKI-aware) -- aware of the rewrite that disassembled his ROS6 setup ~4yr ago (tracks RouterOS dev), even without rebuilding on it
Known
tracks ROS7 template/connection + /routing/filter rewrite
Run
no current deployment
Transport & Packet Typesknow 3 Confident→run 2 Aware
BGP = TCP/179, neighbors MANUALLY configured — no auto-discovery. The defining BGP trait.
iBGP / eBGP (Stub & Non-Stub)know 3 Confident→run 2 Aware
eBGP = inter-AS (prepend + re-advertise); iBGP = intra-AS (no AS-path change + no iBGP->iBGP re-advert -> full-mesh/RR). Residual from supervised ROS6 ~8yr ago
Known
eBGP/iBGP fundamentals (stronger than 'never studied')
Run
supervised ROS6 ~8yr ago, disassembled ~4yr ago
Multi-Hop & Loopbackknow 3 Confident→run 2 Aware
Peer to loopbacks so the session survives a link failure if any path exists. Implies the iBGP-over-loopback + IGP-underneath model.
Route Distribution & Filtersknow 3 Confident→run 2 Aware
Prefix advertise/accept control = routing filter rules (in/out chains) on prefix/AS_PATH/communities. Configured these in ROS6 BGP.
Best-Path Selectionknow 3 Confident→run 2 Aware
local-pref = outbound (internal to the AS); inbound TE = MED (ignorable) / AS-path-prepend / more-specifics, only a HINT (the neighbor's local-pref overrides) -- knows the real asymmetry
Known
local-pref vs MED/prepend TE asymmetry
Run
residual supervised only
Prefix Attributesknow 3 Confident→run 2 Aware
AS_PATH as a core path attribute. BUT ceiling — missed the NEXT_HOP-unchanged-across-iBGP gotcha (chose iBGP-rewrites-to-loopback; real = next-hop stays the external peer → next-hop-self needed). Knows the attributes, not the deep iBGP-propagation subtlety.
Communitiesknow 2 Aware→run 1 Learning
EDGE: BGP communities 'no idea' (NO_EXPORT/policy-tags never used); route-reflector originator-id/cluster-list never touched. The carrier POLICY/scale depth beyond his basic supervised setup
Known
communities 'no idea'; RR scale-policy beyond a basic setup
Run
no carrier-scale deployment
MPLS Basicsknow 2 Aware→run 1 Learning
MPLS forwards by a short label swapped hop-to-hop, not by IP-prefix lookup. The one MPLS concept he genuinely holds at headline level — 'I just heard MPLS delivery differs from classic routing'.
LDP-Based VPLSknow 2 Aware→run 1 Learning
conceptual headline only (LAN-over-MPLS); rides the blind MPLS -> necessarily thin, zero hands-on
Known
conceptual-headline only, rides the blind MPLS
Run
zero hands-on
Bridge Split-Horizonknow 2 Aware→run 1 Learning
Knows L2 bridge split-horizon cold (MTCSWE interface-horizon Teach) but NOT its VPLS application (PE split-horizon across pseudowires to prevent loops in the full-mesh of PWs). Adjacent-knowledge, MPLS-context blind.
L2MTU & Fragmentationknow 2 Aware→run 1 Learning
Strong general MTU/L2MTU mastery (MTCSWE + MTCRE Teach) but NOT the MPLS-specific label-stack-overhead-on-MTU specifics (each label = 4 bytes; VPLS/L3VPN stacks). Adjacent-knowledge, MPLS-context blind.
BGP-Based VPLSknow 2 Aware→run 1 Learning
VPLS-concept GENUINE: picked 'multipoint L2 Ethernet bridge over MPLS' against HARD distractors (L3VPN-VPNv4 / VRF / RSVP-TE-LSP). Knows VPLS=L2 vs L3VPN=L3. The BGP-signaled-VPLS control-plane specifically not probed → concept Aware, mechanism Learning.
L3VPN / BGP L3 Tunnelsknow 2 Aware→run 1 Learning
L3VPN-concept GENUINE: picked 'per-customer VRFs + Route-Distinguisher makes overlapping prefixes unique in MP-BGP' against HARD distractors (RSVP-TE/VPLS/NAT). Holds the overlapping-address-isolation concept. The mechanism depth (RT, VPNv4 NLRI) is blind.
Traffic Engineering Conceptsknow 2 Aware→run 1 Learning
Concept headline only: 'MPLS-TE can consider bandwidth/saturation; I guess it allows bandwidth reservations somehow; I'm blurry here'. Has the TE-steers-by-bandwidth idea, not the machinery.
Bandwidth Allocation & Limitationknow 2 Aware→run 1 Learning
From the TE-concepts probe he holds the 'TE reserves bandwidth' headline but not the allocation/limitation mechanics (per-LSP reservation, available-bandwidth accounting in the TE-DB).

The warrant

No third-party stamp underwrites this page — and that absence is itself the signal. The carrier credential is still ahead of the work, not behind it:

MTCINEInter-Networking Engineer — in progress / not yet taken

A frontier is defined by the solid ground beside it. The honest neighbours:

The questions

Do you run BGP or MPLS in production?

No. I am concept-aware here, not deployment-proven. I ran a supervised RouterOS 6 BGP setup years ago — basic peering, filters — and the BGP rewrite in RouterOS 7 has since disassembled that setup entirely, so even that residue is dated. What I can place truthfully: an autonomous system is one routing-policy domain identified by an ASN, BGP rides TCP/179 with manually configured neighbours, and AS_PATH is both the path record and the loop-prevention. The mechanics I have not deployed at scale — route reflectors, confederations, communities-based policy — read “no idea” in my own assessment. In RouterOS 7 the old v6 instance/peer menus are gone, replaced by /routing/bgp split into connection, template and session menus (with a documented v6→v7 config-conversion step); a connection can even derive the remote AS dynamically from the peer’s first OPEN message. I track that redesign; I have not rebuilt on it.

How deep does the MPLS and VPLS knowledge actually go?

Headline-level. I hold the one-line concepts: MPLS forwards by a short label swapped hop-to-hop rather than an IP-prefix lookup; VPLS is a multipoint L2 Ethernet bridge over MPLS; an L3VPN uses per-customer VRFs plus a route-distinguisher to keep overlapping prefixes unique. Below those headlines it goes dark fast — LDP, penultimate-hop popping, the VPLS control word, RSVP traffic engineering and CSPF all returned “no idea” or “black box” when probed, with zero hands-on. Publicly: in RouterOS 7, MPLS/LDP lives under /mpls/ldp and is not self-standing — it needs an IGP (OSPF or IS-IS) advertising loopback /32s as the underlay before any label is distributed. That underlay — single-area OSPF — is the one part of this stack I do run; the label layer on top of it is the frontier.

Could you build an RSVP-TE traffic-engineering tunnel on RouterOS?

No. I have run zero TE tunnels; RSVP-TE and CSPF are concept-only for me, with no hands-on. The mechanism I can place: in RouterOS 7, beyond the basic LDP label-switched path, traffic-engineering tunnels are signalled by RSVP-TE to build explicitly routed LSPs, and their route is chosen by CSPF — constrained shortest-path-first, where the constraints are interface properties and available bandwidth, with OSPF extensions carrying those TE attributes. Per MikroTik’s own MPLS documentation, RouterOS supports the TE tunnel head-end, explicit paths and CSPF selection, but does not yet implement TE fast-reroute or link/node protection. I can hold the distinction between a label-switched path and an engineered one; standing one up against a live topology and a bandwidth budget is past my deployed edge — I would lab it before I bill it.

Carrier networking on your roadmap?

If the work is BGP at scale, MPLS, or VPLS, it sits past my deployed edge — and I will either lab it before I bill it or point you to a carrier specialist. The IGP underneath it, I already run. Either way, you hear where my line is before the work starts, not after.

Start the conversation