19:46:38 <bblum> talked with paul about cgroups more
19:46:54 <bblum> the first suggestion is being able to move subsystems between hierarchies
19:47:12 <bblum> so cgroups can be mounted multiple times, each with different subsystems attached
19:47:37 <bblum> and this would involve a feature that would let on-the-fly "remove subsystem from the hierarchy" or "add subsystem to the hierarch"
19:47:52 <bblum> this would involve lots of tricky fiddly stuff and is pretty open-ended
19:48:01 <bblum> option 2 is subsystems as modules
19:48:41 <bblum> which seems pretty straightforward as far as implementation goes; the hardest part will involve figuring out how module stuff works in the kernel, and what actually wants to happen
                 within cgroups to support "this subsystem isn't actually here yet";
19:48:51 <bblum> also possible issues when it comes to unloading modules
19:49:13 <bblum> and of course a large chunk of documentation, as it'll be implementing a new interface entirely on top of subsystems
19:49:18 <bblum> i personally like this idea a lot
19:49:40 <bblum> option 3 is a lot smaller and probably could happen if "we picked something too small, let's throw on another thing to work on on top of it"
19:49:45 <bblum> and it would be subsystem dependencies
19:50:02 <bblum> "i, subsystem foo, cannot be attached to a hierarchy unless subsystem bar is also attached"
19:50:22 <bblum> let me know what you think