BSS/OSS Academy
πŸ”§
Section 11.7

Network Orchestrator Variants & Specialised Domains

A taxonomy of the six system classes commonly called "orchestrators" β€” EMS, NMS, domain orchestrator, SDN controller, specialised provisioning engine, and manual / scripted β€” and the COTS products that fulfil them. Maps the specialised and legacy domains every Tier-1 estate also runs (microwave, satellite, SDH, ATM, GPON, SIP/voice, legacy optical, enterprise WiFi) to their typical orchestrator class and answers honestly whether each is reachable from a modern domain orchestrator. Closes with catalog and integration implications, and the anti-patterns that derail programmes which assume the modern domain-controller model applies everywhere.

Sections 11.3 to 11.5 walked through three modern orchestration domains β€” IP/MPLS, mobile core, fixed access β€” where domain controllers, model-driven APIs, and standard YANG service models exist and are reasonably mature. Section 11.6 composed across them. This section addresses everything else: the specialised, legacy, or technology-bound orchestrators that every Tier-1 estate also operates, and which rarely fit into a clean catalog-and-controller story.

The reason this matters is decision literacy. A transformation programme that assumes the modern domain-controller pattern applies everywhere will be ambushed the first time it touches microwave backhaul, satellite SD-WAN backup, an SDH leased-line tail, or a voice/SBC stack. Each of these domains has its own orchestrator class, its own COTS products, and its own integration shape β€” and most of them predate model-driven management entirely.

The Orchestrator Type Taxonomy

Before mapping products to domains, the categories themselves need to be made explicit. "Orchestrator" is used loosely in industry to describe at least six distinct classes of system, each sitting at a different layer above the network and offering a fundamentally different abstraction.

Six classes of system commonly called "orchestrators"

TypeWhat it doesWhere it sits
EMS β€” Element Management SystemManages one vendor's product family β€” config, fault, performance β€” typically with vendor-specific protocols and data modelsLowest layer above devices; one per vendor product line
NMS β€” Network Management SystemMulti-element / multi-vendor network view, usually fault- and performance-centric; weak on service-level abstractionsAbove EMS; aggregates fault/perf, sometimes config
Domain OrchestratorService-aware, single domain, exposes modern APIs (NETCONF/YANG, gNMI, TMF); carries service-level state and intra-domain rollbackAbove NMS β€” Cisco NSO, Nokia NSP, Juniper Paragon, Blue Planet sit here
SDN ControllerProgrammatic configuration of an SDN-managed domain (transport SDN, DC fabric, WAN) β€” usually one technology, one control planeDomain-specific; often embedded within or below a domain orchestrator
Specialised Provisioning EngineNiche technology (SatCom NMS, voice/SBC manager, microwave manager) β€” typically vendor-bound, often pre-NETCONF, with proprietary northbound APIsSide-stack β€” integrates via custom adapter, rarely fits the standard pattern
Manual / scriptedLegacy gear with no modern API β€” provisioning happens via ticket, CLI, spreadsheet, or one-off scriptsTicket-driven; runs alongside everything else and rarely goes away
The definitional boundary
The line between NMS and Domain Orchestrator is the most important one. An NMS knows about elements and faults; a domain orchestrator knows about services β€” VPN instances, slices, subscriber profiles β€” and exposes lifecycle operations on them. Vendors regularly market NMS products as "orchestrators"; the test is whether they carry service-level state and expose service-level APIs, or whether they are still device-centric underneath.
Manual / scripted is not a transitional state
"Manual / scripted" is permanent in real estates. Some legacy platforms (multi-decade SDH circuits, legacy ATM PVCs, certain regulated services) will never get a modern orchestrator because the cost of adapter engineering exceeds the residual revenue. Treating manual provisioning as something that will "go away after the migration" produces transformation plans that never close.

Specialised & Legacy Domains: Typical Orchestrator and Products

The IP/MPLS, mobile core, and fixed access sections covered the modern flagship domains. The table below covers the non-IP/MPLS, non-3GPP, non-PON technologies that real Tier-1 networks also run β€” and that transformation programmes routinely under-estimate. For each, the typical orchestrator class is named, the COTS products commonly seen are listed, and the question "can a modern orchestrator like Cisco NSO talk to it?" is answered honestly.

Likely orchestrator per specialised / legacy product class

Product domainTypical orchestrator classCommon COTS productsNSO-able (model-driven northbound)?
Microwave radio (cellular backhaul, enterprise links)Vendor microwave EMS / NMSEricsson MINI-LINK Manager, Aviat ProVision, Ceragon NMS, Nokia Wavence (formerly 9500 MPR), SIAE-MICROSometimes β€” newer kit ships NETCONF NEDs; older Cu/Eth-radio gear remains SNMP/CLI-only
Satellite (geostationary VSAT, LEO, mobility, backup)SatCom NMS / Service Activation PlatformiDirect Pulse / Velocity, Comtech NMS, Hughes JUPITER, Gilat TotalNMS, Newtec Dialog, ST Engineering iDirect, Starlink Enterprise APIsRarely β€” proprietary northbound APIs; integrated via custom adapter or as an opaque service via partner API
SDH / SONET (legacy TDM transport)Vendor EMS β€” pre-OpenROADM, pre-NETCONFCiena 6500 / OneControl, Nokia 1830 NMS, Huawei iManager U2000, ECI LightSoftMostly no β€” TL1 / CORBA / SOAP northbound; service-level abstractions weak or absent; usually wrapped via adapter or left manual
ATM / Frame Relay (residual legacy data)Legacy EMS / scripted CLICisco MGM, Lucent / Alcatel Navis, vendor CLI, in-house Perl/Python scriptsNo β€” manual or scripted; typically read-only inventory at best
GPON / XGS-PON access (FTTH/FTTB OLT/ONT)Access NMS / domain controller (newer)Nokia Altiplano / AMS, Calix CMS / SMx, Huawei iManager NCE, Adtran Mosaic One, ZTE NetNumenPossible β€” newer access controllers expose REST and TMF APIs; older OLT generations remain SNMP-driven
SIP / Voice / SBC (enterprise voice, SIP trunking)SBC EMS / Voice OSS β€” separate stackOracle ACME (Session Element Manager), AudioCodes EMS / OVOC, Ribbon / Sonus EMS, Cisco CUCM, BroadSoft / BroadWorksSpecialised β€” separate orchestration domain; rarely integrated into a unified XDO; voice keeps its own provisioning chain
Submarine / DWDM long-haul (legacy generations)Vendor EMS β€” pre-OpenConfig opticalCiena OneControl, Infinera DNA / Transcend, Nokia 1830 NMS, Huawei iManagerSometimes β€” newer ROADM platforms speak OpenROADM/NETCONF; older lines are EMS-only
WiFi enterprise / hotspot orchestrationVendor controller / cloud-managed NMSCisco Meraki Dashboard, Aruba Central, Juniper Mist, Ruckus SmartZoneCloud APIs available β€” typically integrated via REST adapter, not via NSO NEDs
The pattern hidden in the table
The "NSO-able" column is the architectural truth. Where a domain has converged on NETCONF/YANG (modern IP/MPLS, modern PON, modern optical), a domain orchestrator owns it cleanly. Where it has not (satellite, voice, microwave, SDH, ATM), every transformation has to choose: build a custom adapter, treat the domain as opaque (call a partner API), or keep it manual. There is no fourth option.

What This Means for the Catalog and Integration Layer

A multi-domain estate that includes any combination of these specialised domains forces specific design decisions on the catalog and the cross-domain orchestrator. Pretending otherwise produces catalogs that cannot represent real services, and orchestrators that cannot fulfil the orders the catalog accepts.

Model the orchestrator class as resource metadata
Each technical service (RFS) should declare the orchestrator class that fulfils it (domain orchestrator, EMS, specialised engine, manual). The cross-domain orchestrator routes to a different adapter shape per class β€” TMF640 to a domain orchestrator, custom REST to a specialised engine, ticket creation for manual. Hard-coding "every domain is NETCONF" forces the orchestrator to lie about what it can actually do.
Manual fulfilment must be a first-class path
If 5–15% of orders touch a manual domain, the orchestrator must model "ticket dispatched, awaiting human" as a real lifecycle state β€” not a special case. Every Tier-1 has a manual residue; orchestrators that ignore it produce orders that "succeed" while the customer-facing service is not actually live.
Keep voice and satellite as separate composition partners
Voice and satellite stacks are usually not worth pulling into the unified XDO. Treat them as composable partner services β€” the XDO consumes their APIs at the service level (activate SIP trunk, activate VSAT terminal) and does not try to drive their internals. This is how every working multi-vendor estate ends up structured in practice.
The catalog must abstract the medium, not enumerate it
A "100 Mbps internet" service may be delivered over fibre, microwave, satellite, or hybrid bonded paths. The catalog should expose the service; the resource layer chooses the medium and orchestrator. Catalogs that hard-code the medium force a new product entry every time a new bearer becomes economic β€” and force every order to commit to a delivery path before site survey.

Anti-patterns Specific to Specialised Domains

Specialised-domain anti-patterns and their failure modes

Anti-patternWhat it looks likeWhat breaks first
"NSO does everything"Programme assumes the domain orchestrator will absorb microwave, satellite, voice, and legacy in a future phasePhase 2 reveals SatCom and voice have no NETCONF/YANG path; adapter cost balloons; either schedule slips or those domains are quietly excluded from the orchestrator scope
Catalog enumerates the mediumSeparate product offerings for "Internet over Fibre", "Internet over Microwave", "Internet over Satellite"Customer chooses Fibre; site is unreachable by fibre; order fails commercially instead of being re-bound to a different bearer at the resource layer
Treating manual as "to be eliminated"Roadmap shows manual domains being orchestrated by Q4; current architecture has no first-class ticket-and-wait stateManual residue persists indefinitely; orchestrator marks orders complete that are still queued in someone's inbox; assurance has no way to detect
Voice swallowed by general orchestratorProgramme tries to orchestrate the voice/SBC stack from the same XDO that runs L3VPN β€” same models, same adaptersVoice service semantics (DID, dial plan, codec, regulatory routing) do not map to L3VPN-shaped models; voice team builds a parallel stack anyway
Treating an EMS as a domain controllerNorthbound to Ciena CMS / Huawei iManager treated as if it were Cisco NSOService-level abstractions absent; orchestrator ends up sending CLI through the EMS; first feature change in the EMS breaks the integration; rollback impossible
Ignoring partner APIsLEO satellite or third-party voice provider treated as a black box with no orchestration integration at allManual provisioning at scale; service activation lead times measured in days; assurance feedback non-existent; transformation has no way to demonstrate end-to-end automation

Section 11.7 Key Takeaways

  • "Orchestrator" is used loosely for at least six distinct system classes β€” EMS, NMS, domain orchestrator, SDN controller, specialised provisioning engine, and manual / scripted. Each has a different abstraction, a different API surface, and a different integration cost.
  • The line between NMS and domain orchestrator is the architecturally important one: service-level state and service-level APIs, or device-centric underneath.
  • Every Tier-1 estate runs specialised and legacy domains alongside the modern flagship domains. Microwave, satellite, SDH, ATM, voice/SBC, and older optical all live outside the clean NETCONF/YANG world.
  • The "NSO-able" question has only three honest answers per domain: yes (modern domain controller), partial (custom adapter + accepted gaps), or no (manual or partner API). There is no fourth option.
  • The catalog should declare orchestrator class as resource metadata, model manual fulfilment as a first-class lifecycle state, and abstract the transmission medium rather than enumerating it.
  • Voice and satellite are usually not worth pulling into the unified XDO. Treat them as composable partner services consumed at the service-level API.
  • The fastest way to derail a transformation is to assume the modern domain-controller pattern applies everywhere. The specialised-domain footprint is what determines whether a programme delivers automation or only looks like it does.