John Jenkins

The MICA paper goes into hardware-assisted affinity management in the context of key-value stores (https://www.usenix.org/conference/nsdi14/technical-sessions/presentation/lim). I'm sure there are plenty of other examples in the literature we can take inspiration...

Most of that does make sense, thanks. Based on a quick look (https://ofiwg.github.io/libfabric/), OFI does have tag matching. You probably are already thinking this, but it seems like the matching...

You could store e.g. the "root" bulk handle as a pointer from the handles and have access to all the needed info, though there's probably a cleaner integration of the...

The use case for de-registration I'm thinking of is a long-running set of mercury processes by which services come and go, i.e. the class ends up getting re-used. De-registering allows...

Preferably not - ideally, with this option stdout would be completely unused by ROSS (with stderr being used for warning/error reporting).

hackathon - people want to post-process output via tw_output and piping. Suppressing ross output would be very helpful in this case.

Looking more in-detail about this, this optimization would actually break a significant fraction of CODES. In particular, some models funnel messages between intermediary LPs, relying on oversubscribed event memory sizes...

I think there might have been a bit of a misunderstanding here... the ticket was discussing was the sending of events between MPI ranks. Some events (particularly for our modelnet...

On the modelnet connection - we essentially increase the global message size by 2x to allow piggybacking modelnet user events with modelnet events. In reality, for the performance-sensitive transports (dragonfly,...

As it pertains to the API, Phil's suggestion would require telling ROSS the effective event size on every tw_event_new. In that case, tw_event_new_large would become tw_event_new_sized. Then the size passed...