Define a multi-hub service offering

[choldgraf]

Context

We have several relationships with communities that have moved beyond the "single hub" approach that our pricing and service model has followed so far. For example, Pangeo, U Toronto, Carbon Plan, etc.

However, our current pricing model doesn't have a natural way to scale to N+1 hubs, and also doesn't reflect the fact that we gain economies of scale in the complexity / maintenance burden for each successive hub on a cluster. Moreover, the type of organization/community that would benefit from a "multi-hub service" is likely different from one that would benefit from a single hub, and this may necessitate a different set of services, pricing structure, etc.

Proposal

We should define a new service model that is geared towards a community with enough complexity or size that they'd want multiple hubs. It's not yet clear what this should be, so here are a few thoughts that @yuvipanda and I discussed for reference:

Pricing

• It would likely begin at a significantly higher cost than our current single-hub offering
• But it would scale sub-linearly so that each successive hub was significantly less

Use-cases

There are likely a few ways that hub features would be split between educational and research use-cases. It might mean that there are actually 2 different service offerings here, or different feature flags within each, but just noting them for now to help us think.

• Authentication service
  • Education: Usually want the same auth service for each hub because they use some institution-wide auth.
  • Research: May want flexibility in auth services, because research communities have more diverse workflows (e.g., some like github, others want institution addresses, etc)
  • Both: Might benefit from having the notion of a hub-agnostic user identity.
• Single- vs. multi-cluster
  • Education: Likely want all of their hubs on a single cluster since they are largely not constrained by access to datasets that are not split across datacenters
  • Research: Likely want access to computing on multiple datacenters if their workflows are constrained by the location of big datasets. (e.g., Launch user sessions in multiple cluster from a single hub features#7)
• Payment structure
  • Education: There is likely a single institution paying for services that a variety of sub-communities (classes etc) use. They'd likely do this via standard "institutional procurement" practices.
  • Research: They likely have a more heterogeneous and decentralized pool of funding sources (e.g., a large community where a subset of PIs have grants that could help fund the service at any moment in time, potentially ear-marked for a subset of users).

Updates and actions

• Do some customer segmentation to figure out how this service might map onto specific stakeholders we'd want to serve.
• Prototyping this here for UofT: https://github.com/2i2c-org/meta/issues/383

References

related issues:

• Formula for calculating hub-specific cloud costs on a shared cluster infrastructure#730
• Strategy for splitting user environments into multiple images infrastructure#935
• Define a process for helping communities with image changes infrastructure#1145
• [Discuss] Provide community guidance and deeper support for some communities #297