Home
Describing Cluster Infrastructure as a Graph
Modern AI systems, comprising diverse scale-up and scale-out interconnect topologies that integrate complex heterogeneous components, connected together via diverse means, face a lack of standardized overall infrastructure description. This hinders benchmarking, simulation, and emulation. infra.proto introduces a graph-based schema to describe AI/HPC infrastructure.
To create an infrastructure as a graph one can use messages from infra.proto. The messages allow a user to easily create logical infrastructure as vertexes and edges. This can be scale up further to describe a massive infrastructure without duplicating content.
infra.proto is a Protocol Buffers (Protobuf) message that serves as the core data model for defining and designing the infrastructure. Protobuf is a schema-based, strongly typed format that models data as messages with defined fields and types in a .proto file. This schema enables efficient serialization and deserialization of structured data.
While Protobuf itself uses a compact binary format, protobuf data can be converted to human-readable formats like JSON and YAML for easier inspection, configuration, and integration with text-based tools. This README uses YAML for its readability and suitability for configuration and data serialization tasks. YAML's clear syntax for nested structures and lists complements the Protobuf-defined data model.
Features
Cluster Infrastructure as a graph is a developing specification informed by practical use cases. The model outlines the following elements for defining infrastructure:
- Device definitions, enabling representation of device internals as a graph
- Device Components , which let users specify internal parts such as:
- nic
- ports
- npus
- Links defined for:
- components interconnect
- device interconnect
- Defining the bandwidth of the links
- Connections relating to:
- internal connections within a device
- connections between devices
A detailed explanation of the topology model is available, including information on its structure, principal components, and its application in design and analysis. This resource presents guidance on the principles of topology and effective application methods.
Annotation
This section explains how to annotate infrastructure with details such as DeviceType and Rank Identifier. It includes a model description and examples of linking physical attributes to the logical infrastructure definition.
Getting Started With Topology Creation
This walkthrough guide demonstrates how anyone can create a topology from scratch, highlighting key steps and best practices to build a solid foundation. It offers a clear, step-by-step approach that makes topology creation accessible to beginners and experts alike.
Community
Use our community resources to get help with Infrastructure As A Graph: