Key Features

Dayu covers the complete lifecycle of a cloud-edge stream analytics application: DAG composition, runtime deployment, scheduling, execution, observation, and extension.

DAG-based AI service pipelines

Applications are modeled as directed acyclic graphs rather than fixed linear chains. A DAG may branch into several AI services, join their results, and assign each logical stage to a different execution node. Workflows can be created in the frontend or imported and exported as .dag files.

Coordinated cloud-edge runtime

Dayu deploys the services required by an application as a coordinated runtime and makes the workflow available after that runtime is ready. Controlled updates keep in-flight work consistent when deployment decisions change.

Adaptive scheduling and deployment

Pluggable policies cover source selection, deployment, data configuration, offloading, and admission. Built-in baselines and research schedulers can be compared under the same runtime and observability model. See Customize Scheduling for the policy catalog and extension points.

Hook-driven extensibility

Hooks and templates let developers extend scheduling, data generation, processing, monitoring, and visualization without rewriting the runtime lifecycle.

Built-in AI service catalog

The catalog includes detection, tracking, classification, recognition, model switching, and structured traffic analysis. Generic input/output forms keep these services composable in different DAGs.

Observability and experiment feedback

Result visualizers render frames, structured overlays, curves, topology, CDF, and Gantt views. System telemetry covers compute, memory, network, model execution, scheduling, and queue behavior. Result records and system snapshots can be exported for offline analysis.

Next steps

Read the architecture overview, follow getting started, or use the developer guide to extend Dayu.