Configure Shared Cache for Upstream Models
Overview
Configure the shared cache for upstream models in order to share models with other organizations having Istari installations.
The shared cache for upstream models is an object store, configured similarly to other Istari object stores as described in the Object Store Configuration section.
Shared Cache for Upstream Models vs. Istari Tenant Object Store(s)
Important differentiations for the object store(s) configured as a shared model cache include:
- Shared caches for Upstream Models should be provisioned and managed by the upstream Istari installation administrator, with appropriate permissions granted to the downstream Istari instance and downstream end users for read access to the shared cache.
- The shared cache for an upstream organization may be configured to isolate storage containers on a 'per downstream' basis (e.g. one object store cache per downstream), or, a single shared cache may be configured to partition a single shared cache for use by multiple downstream organizations.
- The configuration of the shared cache for upstream models is the responsibility of the upstream organization (the owner of the models to be shared).
Shared Cache Configuration Options and Considerations
Note: Configuration choices and configuring access to the shared cache is the responsibility of the upstream organization, and these choices and configuration should be carefully considered and reviewed by network and security administrators prior to providing access to downstream organizations.
- Isolated Shared Caches (one per downstream organization involved in remote sharing)
- This option involves:
- Provisioning a separate cache object store for each downstream identity created for sharing
- Configuring access to that storage container so that the downstream Istari installation can write to the container, and downstream end users can read from the container
- Single Shared Cache for Upstream Models
- This option involves:
- Provisioning a single cache object store, for example, a single S3 bucket, to be used as a shared model cache by all downstream organizations
- Partitioning that cache object store to ensure each downstream organization has a separate, isolated folder or partition
- Configuring access to each object store partition so that the downstream Istari installation can write to the cache, and downstream end users can read from the cache, according to the different identities and access tokens (or 'PAT's) created for each sharing relationship.
- For the choice of using an S3 bucket, considerations could include the use of an S3 Access Point, or other IAM practices.