ruote tmp/log_2012-12-22.html

2012-12-22 00:49:00 utc lmarburger_ jmettraux: ha! i didn't even bother to follow through to see the update. you made a big deal about it :p
2012-12-22 00:49:29 utc jmettraux yes, since I was quite wrong
2012-12-22 00:57:18 utc mburnett hey jmettraux
2012-12-22 00:58:28 utc mburnett thanks for the interesting await blog post
2012-12-22 00:59:00 utc mburnett typedef and I have been working this month on a ruote-based replacement for our current workflow system
2012-12-22 01:02:27 utc mburnett we should have some benchmarks that you will find interesting
2012-12-22 01:04:09 utc jmettraux mburnett: hello, looking forward to them
2012-12-22 01:04:16 utc mburnett :)
2012-12-22 01:06:58 utc mburnett one thing is that we have noticed that with trackers (await in our case), every message loops over every tracker defined in every process
2012-12-22 01:08:05 utc jmettraux yes, as some trackers are "engine wide"
2012-12-22 01:08:18 utc mburnett ic
2012-12-22 01:09:07 utc mburnett the concern is that this makes the process runtime lower bound O(num_processes * avg_ops_per_process * avg_awaits_per_process)
2012-12-22 01:10:39 utc mburnett i guess i should say expected run time rather than lower bound
2012-12-22 01:15:23 utc jmettraux indeed
2012-12-22 01:18:14 utc jmettraux I have a storage implementation that keeps trackers only per-process (no engine-wide trackers)
2012-12-22 01:19:29 utc jmettraux but I'm not sure how I would generalize those local process only trackers... placing them in a variable in the root of the process? Yes, but what happens when the root becomes unreachable
2012-12-22 01:19:40 utc jmettraux (as is the case for some forgotten branches...)
2012-12-22 01:19:58 utc jmettraux well I guess I could copy over the variables for the isolated branch
2012-12-22 01:20:52 utc jmettraux It's doable, but it's not a one hour hack
2012-12-22 01:20:52 utc mburnett i expected improvements to tracker performance would not be trivial :D
2012-12-22 01:21:23 utc mburnett i'll probably have a pdf to mail you with lots of info about what we're doing late next week or early the following
2012-12-22 01:21:36 utc mburnett we just need to do some more measurements, and get the pdf ok'd (no patient info)
2012-12-22 01:22:04 utc jmettraux ok, it would be really nice if that info could be shared on the mailing list, I'd understand if you want to keep it private though
2012-12-22 01:22:41 utc mburnett i'll ask about that
2012-12-22 01:22:59 utc mburnett i bet they will approve it, we try to be open
2012-12-22 01:23:05 utc mburnett if not for hippa, it would be easy :)
2012-12-22 01:23:30 utc mburnett i talked this morning with my manager, and his response was "regardless of how we proceed, we want to give the ruote community/author back whatever info we can"
2012-12-22 01:29:24 utc jmettraux thanks!
2012-12-22 02:51:54 utc mburnett jmettraux: we're trying to profile ruote with a particular process, but it appears that the actual worker is running in a separate process, so the profiler is giving useless results, any tips?
2012-12-22 03:16:13 utc jmettraux mburnett: run the profiler in the process that contains the worker?
2012-12-22 03:16:24 utc jmettraux or run everything in the same process