As a Stage 1 project in LF Edge, Project Alvarium aims to build a framework and SDK for trust fabrics that deliver data from devices to applications with measurable confidence.

See the vision at:

Project Alvarium at a Glance

  • New LF project forming to focus on system-level trust and data confidence
  • Differentiated in its comprehensive view and in delivering data to applications with measurable confidence
  • Unifying, not reinventing trust insertion technologies
  • Relevant to all markets and solution stacks
  • Seeded by Dell Technologies code

Project Mission:

  • Create the framework and open APIs that bind together existing open source and commercial value-add for trust insertion, develop confidence score algorithms
  • Collaborate with other LF projects and industry efforts (OSS, SDO) to unify existing and emerging trust insertion technologies and refine scoring algorithms

What is a Data Confidence Fabric (DCF), or “trust fabric”?

  • A Data Confidence Fabric (DCF), or more generally-speaking trust fabric, is a virtual overlay that aids in the delivery of data from devices to applications with measurable trust characteristics.
  • A DCF is a loosely-coupled collection of various trust insertion technologies, bound together with an open framework
    • Example technologies include tools for silicon-based Root of Trust (RoT), open authentication and data ingestion APIs, metadata handling, immutable storage and blockchain/ledger
  • The Alvarium framework features open APIs and integrated algorithms to generate confidence scores for data based on the trust insertion technologies used and overall context
  • There is no single DCF, rather each entity/organization can build their own fabric with preferred technologies using the Alvarium framework
    • A trust fabric built with widely trusted ingredients will naturally produce the highest data confidence scores
    • Confidence scores normalize across systems of systems as data flows through intersecting trust fabrics
  • Key differentiation from other efforts focused on security, privacy and trust:
    • Holistic, system-level focus
    • Confidence scores to enable organizations to act with measured risk based on policy appropriate for the use case / context, working across heterogenous systems of systems

Why we need to collaborate on a global trust fabric

  • Pervasive sharing and monetization of data, resources and services across heterogenous systems of systems spanning public and private boundaries
    • Can also include trusted sharing/exchange of data sets for training AI model
    • The common “zero trust” model isn’t scalable, access policy needs to be attached to trustworthy data
  • Consolidating workloads on common infrastructure in a trusted fashion
    • Enable sharing of data/services based on policy while protecting privacy and IP
    • Address common debates on data ownership
  • Meeting compliance requirements (e.g. GDPR) at scale
    • Enables organizations to trigger deletion of distributed data in place when a user requests to revokes privacy consent

Example end-to-end trust insertion points

Example OSS trust insertion technologies

Initial DCF prototype

  • Dell Technologies’ initial Data Confidence Fabric (DCF) prototype (completed in August 2019) demonstrated a trust fabric comprised of a mix of open source and commercial technologies
  • Prototype was deployed entirely on one edge system to locate policy insertion for data monetization/compliance as close as possible to the data source
  • Alvarium framework unifies the various loosely-coupled trust insertion elements
  • Solution could just as well be deployed in a distributed fashion
  • Next steps in prototyping – demonstrating technology swapability, for example exchanging Project Concord ledger for Hyperledger or IOTA
  • Dell will contribute the Alvarium framework code to seed the project

Example Confidence Scoring

  • Scoring creates a weighted confidence depending on trust insertion technologies implemented in a given trust fabric
  • Dell’s initial DCF prototype leveraged a simple linear scale for simplicity
  • Scoring algorithms will require industry collaboration to develop
  • Initially via OSS but may require some standards work
  • Likely to make sense for weighted scoring, some factors that zero out confidence

  • No labels