Arvados

Thoughts on what direction to go to make people more excited about Arvados when they read it.

I think maybe one place to start is who the audience is. Tentatively I'm going to suggest writing for people who want to learn a little bit more about the capabilities of Arvados and are technical but not familiar with the Arvados architecture. So then the goal of getting into technical details is to pull those people in (so it doesn't read like a fluff piece that asserts things without backing it up) without being overwhelming.

So perhaps instead of talking about 2.7 features specifically, I think we could start by talking about the kind of scale Arvados is designed for and being a little more specific about uses, e.g. our customers need to run a workflow that takes 4-6 hours on a batch of 2000 samples, and they want to get the results the next day, that's not an easy thing to do, but Arvados can do it. Then we can examine several specific approaches that we've had to develop to let us achieve that.

We should also try to provide more specific numbers -- the attached graph doesn't have any identifying information.

The section on container logs is pretty good, but I think we can take out the comparison to the old system and just keep the part where we talk about how we don't waste time transmitting logs that don't get viewed.

I think the section on request prioritization is also good.

Unfortunately talking about client request limiting and retries doesn't seem very exciting. Maybe replace it with new section(s) about arvados-dispatch-cloud and/or compute node local keepstore?