search

Project-MONAI / monai-deploy

MONAI Deploy aims to become the de-facto standard for developing, packaging, testing, deploying and running medical AI applications in clinical production.
Apache License 2.0
97 stars 22 forks source link

DRAFT:MONAI Deploy Workloads Guidelines #23

Closed Danulf closed 3 years ago

Danulf commented 3 years ago

Workloads document for review. Note, this was checked in elsewhere directly and this PR also correctly places it in the guidelines folder.

whoisj commented 3 years ago

Honestly, I am not getting any value from this document. It could be just me, and my lack of medical experience might be primary factor is my inability to derive utility. While it's useful to have some of these kinds of workflows defined, I am not able to work backwards from this document to requirements for the software.

I propose that we focus on the generic types of workloads we can support today, and the potential types of workloads we can support in the future. I'll start with an example:

Type Start Complete Availability
Sync Results synchronous synchronous next
Async Results synchronous asynchronous current
Queued w/ Async Results asynchronous asynchronous future
Streaming synchronous synchronous future
Sync Results w/ Preemption synchronous synchronous future
Async Results w/ Preemption synchronous asynchronous future

Are there other workload/workflow types we should consider here?

Once we have this list specified, we can tie requirements to these generic work types. Then, we can add real-world examples for each work type to help clarify their utility.

Does that make sense to anyone else?

dbericat commented 3 years ago

I really liked some of the computational requirements and expected timing on the first 3 or 4 workloads. We should aim for the same level of detail for the rest.

dbericat commented 3 years ago

There is this document we could collaborate on during co-creation:

MONAI Deploy Workloads doc

MONAI Deployment Workloads spreadsheet

My suggestion would be to use the doc as main source. Definitions, naming, etc. happens there. Raw data si captured on the spreadsheet and linked and charts screenshots posted to the doc.

Once Reviewers are good, we comment on this PR and merge it.

Do we agree? @whoisj @ristoh @hshuaib90 @jonanswer @Danulf @GreySeaWolf @CPBridge

whoisj commented 3 years ago

There is this document we could collaborate on during co-creation:

MONAI Deploy Workloads doc

MONAI Deployment Workloads spreadsheet

My suggestion would be to use the doc as main source. Definitions, naming, etc. happens there. Raw data si captured on the spreadsheet and linked and charts screenshots posted to the doc.

The only issue with the spreadsheet is that its not visible, as far as I can tell, by the world in general. We have to commit to converting any data captured in the sheet to markdown to be recorded in this repository. @dbericat, if you agree with that then I agree with collecting the information in the spreadsheet.