07:22 <Yorlik> heller_: These Tony van Eeerd talks are really enlightening. I especially liked the example of these Berkley guys building an OS and having it explode after running for four years

07:24 <Yorlik> The queue he presented made one thing pretty clear for me: The hard part seems to be (at least in my eyes), that lock free stuff it's not just difficult, but also especially hard to reason about - and how can you get something right if reasoning about it is already hard?

07:25 <Yorlik> Maybe we need other thinking or language constructs to really think about it.

07:26 <Yorlik> The acquire-release semantics seem to be a nice thing, but to me it looks as if this is just scratching the surface

07:28 <Yorlik> The example of discussion in the C++ standard commitee about the correctness or non-correctness of not correctness of the code example with mixed acuire-release and relaxed semantics looked like a good example about how much we are still missing in properly understanding concurrency.

07:30 <Yorlik> I'm trying to get my head around by thinking about concurrency in terms of contracts and expectations about data properties: What do I expect from a certain piece of data and what kind of contract exists (or does not exist)

15:42 <jbjnr__> hkaiser: yt?

15:43 <hkaiser> jbjnr__: here

15:43 <jbjnr__> quick question ...

15:44 <jbjnr__> I've forgotten how the parelport bootstrapping happens. I wanted to get it running on the sockets libfabric for debugging thins ...

15:44 <jbjnr__> after each node starts, it must send a message to the agas node

15:44 <jbjnr__> but where and when does this happen.

15:45 <jbjnr__> I init the agas_ locality on the agas node, but the other nodes need to connect to it and I can't remember where they call it from

15:45 <hkaiser> jbjnr__: brb

15:45 <jbjnr__> ok

15:52 <jbjnr__> I think I know what to look for. I remember

15:54 <jbjnr__> (maybe)

16:20 <hkaiser> jbjnr__: that's in the bigbootbarrier code

16:27 <hkaiser> jbjnr__: https://github.com/STEllAR-GROUP/hpx/blob/master/src/runtime/agas/big_boot_barrier.cpp#L439

16:27 <hkaiser> this handles the AGAS incomming requests on locality 0

16:28 <hkaiser> https://github.com/STEllAR-GROUP/hpx/blob/master/src/runtime/agas/big_boot_barrier.cpp#L575

16:28 <hkaiser> this is handling the response from locality on all localities

16:28 <hkaiser> here is the initial message sent to locality 0: https://github.com/STEllAR-GROUP/hpx/blob/master/src/runtime/agas/big_boot_barrier.cpp#L783-L799

16:40 <jbjnr__> thanks hkaiser I'll experiment