4 ms·

This is interesting and I have never really thought about this. Could you expand a little bit more on it? Would the still be stored locally or would the profile

This is interesting and I have never really thought about this. Could you expand a little bit more on it? Would the still be stored locally or would the profile have all of your content (kind of like if your drop box was also your OS). And would these devices need to be always connected to the internet, or would they have capabilities on their own, and just be able to hook up to the cloud for more compute Also how is this different from sshing into a system to do work and running jobs on a Linux box?

hoosieree 4y ago
Those are great questions. You could get 60% of the way there with Dropbox syncing your OS, with a few caveats (screen brightness should be per-device). But if you start drafting an email on your phone, you probably would like to finish writing it and send it from your laptop. Gmail and Outlook do this today, but only because their phone app and their desktop app are working together with a remote server. If I was building a prototype, I'd start with an always-connected model and "eager" state updates, because caching is hard. To support concurrent writers who both want to be able to undo each other's edits suggests using an OT[1] or CRDT[2] as the basic data type, rather than files. Each node regularly advertises its current capabilities (as well as things like impending sleep states or battery life) to a central scheduler which sends tasks to nodes and collects results. Note - a lot of these ideas come from a project by my colleagues called unis[3], but applied to the topic of operating systems. There are some interesting design choices. Does the dispatcher make all the scheduling decisions, or can the user explicitly say things like "it's ok if Minecraft drops frames, but that background rendering job has to be done by 5pm"? Synced settings are not qualitatively different from multiple clients SSHing into a Linux server, but if any client can suddenly turn around and start acting like a server on a whim, that would be cool. [1]: https://en.wikipedia.org/wiki/Operational_transformation https://en.wikipedia.org/wiki/Operational_transformation [2]: https://en.wikipedia.org/wiki/Conflict-free_replicated_data_type https://en.wikipedia.org/wiki/Conflict-free_replicated_data_... [3]: https://github.com/periscope-ps/UNISrt/blob/develop/sphinx/guides/introduction.rst https://github.com/periscope-ps/UNISrt/blob/develop/sphinx/g...

hoosieree 4y ago

Those are great questions. You could get 60% of the way there with Dropbox syncing your OS, with a few caveats (screen brightness should be per-device). But if you start drafting an email on your phone, you probably would like to finish writing it and send it from your laptop. Gmail and Outlook do this today, but only because their phone app and their desktop app are working together with a remote server. If I was building a prototype, I'd start with an always-connected model and "eager" state updates, because caching is hard. To support concurrent writers who both want to be able to undo each other's edits suggests using an OT[1] or CRDT[2] as the basic data type, rather than files. Each node regularly advertises its current capabilities (as well as things like impending sleep states or battery life) to a central scheduler which sends tasks to nodes and collects results. Note - a lot of these ideas come from a project by my colleagues called unis[3], but applied to the topic of operating systems. There are some interesting design choices. Does the dispatcher make all the scheduling decisions, or can the user explicitly say things like "it's ok if Minecraft drops frames, but that background rendering job has to be done by 5pm"? Synced settings are not qualitatively different from multiple clients SSHing into a Linux server, but if any client can suddenly turn around and start acting like a server on a whim, that would be cool. [1]: https://en.wikipedia.org/wiki/Operational_transformation https://en.wikipedia.org/wiki/Operational_transformation [2]: https://en.wikipedia.org/wiki/Conflict-free_replicated_data_type https://en.wikipedia.org/wiki/Conflict-free_replicated_data_... [3]: https://github.com/periscope-ps/UNISrt/blob/develop/sphinx/guides/introduction.rst https://github.com/periscope-ps/UNISrt/blob/develop/sphinx/g...