Open mmccool opened 2 years ago
At this point, the problem appears to be mostly in the proposals. I think what I need to do it look through the document carefully and identify places where terminology is not consistent, and suggest some specific changes in this issue (but not create a PR necessarily until we discuss; I don't want to change someone else's use case or proposal without consultation).
Will make a Draft PR for the purposes of discussion, but let's not merge it until we discuss, because there is more than terminology we have to align here, we also have to talk about scope.
Another terminology issue I have noticed concerns the term "rendering", where there’s a difference between rendering as in “display an image on a screen” and rendering as in “compose an image from different graphical elements, rules, and visual effects” (which is a more compute intensive task).
Sorry I haven't been keeping up on this, but will try to find time soon. Needham's input looks to be generally useful, though.
Some terminology is used inconsistently, but this also raises a few other things we need to better align.
The Distributed Worker Proposal uses "Computing Resources" or "Distributed Computing Resources" to talk about all possible locations where a workload can be run, but some of the proposed standards and the "Seamless code sharing" proposal talk about "Edge Cloud" as the offload target.
It's not clear if Cloud resources are included in the Edge Cloud. However, Cloud in particular is potentially useful as an "offload target", even if only as a fallback option. IMO, as long as a target meets the QoS requirements of the workload, it should be considered as an offload target.
We could tweak these definitions (for example, to explicitly include Cloud resources in the Edge Cloud) and/or could use "Computing Resource" consistently (or another term, like Distributed Computing Resource or Remote Computing Resource, the latter if we do NOT want to include the client, which may be more logical given how most of the text is written).
The other, related issue is that a Computing Resource may be owner-managed (e.g. a service running on a Desktop) and not be an extension of the cloud, per se. As long as a target is discoverable and satisfies the network API and protocol allowing offload it should be considered. Perhaps "Compute Resource/Edge Cloud Management mechanisms" should be added to the out-of-scope section?