Abstraction for Network Interoperability and Automation 

Dave Duggal
Found and CEO @EnterpriseWeb

Originally posted on LinkedIn

Hypergraphs support variety and change to accommodate Telecom complexity and evolution

This article was motivated by two good posts by Larry Peterson and Bruce Davie, which are worth reading – What Happened to Network Automation? and Automation: An Over-Constrained Problem.

It’s hard to argue with these two accomplished, PhDs who have materially contributed to the networking community. They call out issues and frustrations that everyone in the Telecom industry has encountered regarding the struggles to automate network operations. All the custom code and manual point-to-point integration it takes to deliver network services make Telco operations a big ball of mud, and Cloud and AI initiatives get stuck in that quagmire of opaque, siloed use cases.

While Telcos are “over-constrained”, the situation is not hopeless and intractable. Enabling Telco automation, intent-based orchestration and AI initiatives for intelligent and responsive networks is not just desirable, it’s critical to Telco internal transformations. Likewise, for Telcos to enable their enterprise, manufacturing and government customers with AI and edge, they must provide north-bound customer-facing abstractions that simplify and automate the exploitation of infrastructure – otherwise someone else will and telcos will be relegated to bit-pipes. 

Performant, scalable, resilient, high-level abstraction is most valuable beachhead in all of IT – it is the holy grail! Everyone from Jinsung Choi (PhD Computer Science, Chair of the Board at O-RAN, Deutsche Telekom SVP, Head of T-Labs) to Stephen Saunders (MBE, formerly founder and CEO of Light Reading, now Silver Linings, and Vice Chairman Of The Board at Laser Light Communications) is talking about it (Jinsung’s post, Steve’s article). 

“The entire history of software engineering is that of the rise in levels of abstraction.” Grady Booch, Computer Scientist, one of the inventors of UML 

To discuss the solution space, we first need to objectively consider the problem space. Telco operations are inherently complex, but there is also a lot of accidental complexity. This problem requires clear-eyed introspection and holistic systems thinking. Way too many people in the industry equate the presence of standards with “interoperability”, but if that was the case, the issues wouldn’t exist. 

The old joke is “Standards are such a good idea, everybody wants their own”. It’s not that any discrete standard is bad, but there are many standards-bodies with many standards, which has been compounded by the explosion of OpenAPIs. There is no unified standard for end-to-end Telco interoperability and automation. 

The reality is that each telco selectively implements various versions of standards from multiple standards-bodies and mixes in their own concepts to fill gaps and reflect their specific business and regulatory environments. The result is custom Telco-specific information models, which are typically hierarchical, bloated and static, making them expensive to maintain and hard to change. 

These Telco information models are generally not complete or performant enough to actually run real-time operations so each use case ends up being its own custom-coded silo. Use cases incorporate metadata from a few relevant standards, but it is generally Infrastructure-as-Code (IaC), not a common model. Instead, telcos expose use case specific templates that enable telco teams to integrate, compose, deploy and manage vendor functions. The templates generally include standard references and use widely adopted formats (BICEP, Terraform, etc.), but they are really just simple configuration languages with narrow custom DSLs for the use case. You have to use the word “Declarative” with a grain of salt in this context. 

Vendors are generally only partially-conformant to standards. They selectively adopt versions of specific standards just like the telcos do per product. Inconsistent use of standards sets up a mismatch – everything is a snowflake –  that makes onboarding an exercise in forcing square pegs in round holes; which ensures custom one-off integration.

Onboarding is “lossy” as the vendor’s native interface is constrained by the template, proprietary properties and configurations that could be useful are left-out. The process is time-consuming, tedious and repetitive. In effect, the vendor has to build and maintain a custom interface to conform to the template for every use case, for every telco, and then must update their custom interfaces as requirements, technologies and standards change. This is such a burden that vendors have to limit the number of industry proofs-of-concepts they do, because they don’t have the resources. Telcos may believe this is a vendor problem, but of course the cost is reflected in vendor pricing and the constraints on onboarding are rate-limiters of Telco innovation. 

The solution to custom-coded, manually-integrated, tight-coupled silos is not more custom-code, manual integration and tight-coupling. Instead of simply “shifting left”, making already overloaded developers responsible for Ops tasks, the answer is to “shift up” to abstractions that simplify and automate IT. 

I can already hear the grumblings from seasoned network engineers moaning that they’ve heard this story before. Yes, it’s true, historically, model-based solutions have been limited, usually within a domain and typically abstracting low-level infrastructure (ODL, Netconf/YANG, etc.). In effect, anything outside the model ends up being custom coded and manually integrated. However, that shouldn’t be an excuse to give up. The problem is not intractable. It’s not the Kobayashi Maru

The lesson from past failures is that the scope of declarative intent is equal to the breadth and the depth of the model. If customization is the roadblock to automation, then the solution is that the model must also be able to accommodate variance at the object level. Here’s how we do it – 

EnterpriseWeb is a knowledge-driven platform for real-time intelligent automation. At the heart of EnterpriseWeb is a hypergraph, which supports highly-complex, multi-dimensional, conditional and evolving relationships. It’s built to abstract complexity and accommodate change, at scale. It supports the rapid modeling of complex, distributed domains that support global visibility, end-to-end automation and centralized policy control. It provides a single-source-of-truth, a unified operational model that offers shared domain semantics, metadata and state to automate Day 0, 1, 2+ operations. It’s a central source of context for GenAI, intent-based orchestration, and AIOps. 

For telcos, the platform includes a harmonized standards-based Telecom Ontology, ready-to-use, out-of-the-box. The ontology includes standards-based concepts from 3GPP, O-RAN, MEF, TMF, ETSI, TOSCA, etc. It supports multiple versions of standards and can evolve with the industry. Telcos can extend the domain with their own information models so it aligns closely with their specific operations. The result is a “super” SMO that can work across the RAN, core and Transport and between layers 1-7. 

In EnterpriseWeb, onboarding is modeling heterogeneous solution elements as typed objects. On ingest of files (images, artifacts, documentation) or scans of endpoints, our platform performs algorithmic entity extraction and maps the properties, behaviors, constraints, interfaces to the ontology. The type system auto-fills the objects with standards-based metadata and relationships, and wraps the object in a set of standards-based interfaces. 

Importantly, the onboarding process allows for the modeling of idiosyncratic, vendor/product-specific details in the object. It’s an intelligent and flexible template that accommodates variety while driving minimum viable normalization with conformance checking so all elements can be discovered, declaratively composed, and managed via standards-based interfaces. 

At runtime, the platform acts as an intermediary translating standards-based commands for each of the participating elements through their native APIs or CLI. This approach enables Telcos to cope with heterogeneous elements and onboard innovation, without the custom development and glue-code that plagues typical deployments, which makes them brittle and opaque. 

The implements high-level functional programming techniques so it naturally offers referential integrity – every object and every dependency is transparent, navigable, query-able in the platform and every interaction has a real-time trace. Moreover, we implement a log-style, append-only, immutable store so we have a natural version control system with complete temporal history of all changes to all objects.

EnterpriseWeb is a pioneer in Telecom virtualization and automation. It led the first ETSI NFV Proof-of-Concept, “CloudNFV” went on to win 6 consecutive TMF Catalyst awards. Last year the company ran an award-winning Intel 5G MEC testbed demonstrating a “Super SMO” working across the RAN, Core and Transport to maintain low-latency and energy consumption in support of telco edge and sustainability initiatives. Recently, EnterpriseWeb won a Light Reading Leading Lights award for its ground-breaking work on “telco-grade generative AI for intent-based orchestration” in collaboration with Microsoft. With EnterpriseWeb, developers can “talk to the network” to design, deploy and manage complex services. 

The Platform technology is based on agents, hypergraphs and serverless functions – a horizontal architecture for complex event processing. The platform eliminates traditional static, siloed, middleware stacks  (no 3rd party queues, gateways, brokers, integration tools, workflow engines or orchestrators required). The comprehensive platform has an edge-optimized 50MB footprint. It can deploy at the edge providing full GenAI and AI automation solution for secure, low-latency and high-performance processing local to the customer workloads. EnterpriseWeb nodes can connect in an “application fabric”, akin to a service mesh, allowing fleets of edge nodes to be centrally orchestrated and managed. 21 awarded US patents. Additional patents pending. 

Related Links 

Intel MWC24 Demo: “Secure Dev-centric Networking with CAMARA APIs”

Red Hat MWC24 Demo: “GraphOps for the AI-enabled Telco”

Microsoft MWC24 Demo: “Telco-grade generative AI for intent-based orchestration”

Blogs: GraphOps for the AI-enabled telcoSorry Telcos, graph databases are no way to run a business

Interview: Live on TheCube from MWC24 with Red Hat

Article: EnterpriseWeb and Fortinet Demonstrate Adaptive Security for 5G Edge Networks

White paper: Over-the-Top Orchestration with Telco Cloud

White paper: “Above the Clouds” by David Linthicum, Cloud CTO Deloitte