User problem
TauGrid workloads need a coherent way to use durable storage for data such as datasets, checkpoints, shared working state, and outputs. Users should not need to understand or manually assemble all underlying infrastructure details before a workload can reliably read and write its data.
Desired outcome
Produce a design for a managed storage layer within TauGrid.
The design should establish the intended user experience and enough of the system contract to support future implementation planning. It should explain key assumptions, tradeoffs, failure modes, and how the proposal fits TauGrid's existing CLI, APIs, controllers, and workload lifecycle.
A reasonable design will walk through the actual cloud infrastructure used making appropriate tradeoffs as well as the integration/implementation within the TauGrid ecosystem and potentially training a small model showcasing the implementation using a modern framework.
User problem
TauGrid workloads need a coherent way to use durable storage for data such as datasets, checkpoints, shared working state, and outputs. Users should not need to understand or manually assemble all underlying infrastructure details before a workload can reliably read and write its data.
Desired outcome
Produce a design for a managed storage layer within TauGrid.
The design should establish the intended user experience and enough of the system contract to support future implementation planning. It should explain key assumptions, tradeoffs, failure modes, and how the proposal fits TauGrid's existing CLI, APIs, controllers, and workload lifecycle.
A reasonable design will walk through the actual cloud infrastructure used making appropriate tradeoffs as well as the integration/implementation within the TauGrid ecosystem and potentially training a small model showcasing the implementation using a modern framework.