IRC Logs for #io Wednesday, 2014-05-21

*** springbok has joined #io00:01
*** bjz has joined #io02:07
*** gatesphere has quit IRC02:11
*** bjz has quit IRC02:12
*** bjz has joined #io02:19
*** republican_devil has joined #io03:25
republican_devilhows the actor based concurency?03:25
prologicsay what?03:38
*** codestorm has joined #io03:47
robonerdis that planned republican_devil ?04:03
prologicsay what?04:39
prologicIo uses an actor based concurrency model arleady04:39
*** bjz_ has joined #io04:49
*** bjz has quit IRC04:50
republican_devilyes but how does it WORK?05:09
prologiclike the actor model?05:32
prologicit’s an actor model implemtnation05:32
prologicit should work the same as any other05:32
prologicpersonally I prefer event-driven to actor05:32
prologici.e: asynchornous05:32
*** bjz_ has quit IRC05:39
republican_devilseems cool05:39
republican_devilcan it use liek 20 cpu in 1 big box?05:39
*** bjz has joined #io05:40
robonerdprologic why do you prever event based over actor?05:47
prologicwell because I can conceptually model a lot of things with events and components05:49
prologicI’m the author of one such framework for the Python progrmming langauge (
prologicI find event-driven programming easy to understand for concurrency I/O handling and many other applications05:50
prologicit also scales quite well05:50
robonerdwhat problems have you had with actor model?05:50
prologicI’ve actually never practically used it myself05:51
prologicbut one of the major problems I find with Actor based concurrency is with their implemtantions05:51
prologicor perhaps it’s a limitation of our CPU architectures and threading05:51
prologicNB: OS threads05:51
prologicActor Model (IIRC) require  actors to execute simulteneously05:52
prologicAnd whilst this is easy to achieve, it means you end up costing a lot of performance and scalablity in serlaization of data/messages to/from actors05:52
prologicAt least as far as I understand it :)05:52
prologicNo preactical experience with Actor Model though so my comments could be all totally wrong :)05:53
robonerdcheck that out05:53
robonerdseems actor model can also be event driven?05:53
prologicit’s not quite the same05:55
prologicbut similar yes05:55
prologicone of the things I’ve found IHMO is that Actor Model05:55
prologiccan actually be modelled on top of event-driven mdoels05:55
prologicso yeah *shrugs*05:55
prologicI’ve seen a few Actor Model frameworks and implemtnations in Python for example05:55
prologicand as I said, it doesn’t scale all that well due to the cost of serialization05:56
prologicYou end up with a local distirbuted concurrency model05:56
prologicwhere messages between actors are sort of distributed across OS threads (processes)05:56
prologicunless you have real shared memory between procsesse it gets expensive05:56 is impressive05:57
prologicit’s clearly wirtten in some low-level implemtnation as a library to higher level languages05:58
prologicor they’d not get that kind of performance/scale05:58
prologicthing is05:58
prologiccircuits (python framework) with it’s event-driven model can achieve quite good performance and scalability05:58
prologicit’s not written in C, so it’s not going to scale to billions of events per second or anything like that but it’s good enough for most uses :)05:58
republican_devilso why doesnt everyone use m,ultithreaded liek aolerver?05:58
prologicand no compliation required of low level non-cross-platform code :)05:59
prologicthing is everything has it’s pros and cons05:59
prologicsome concurrency models are better suited to other tasks05:59
prologicbut a lot of concurrency models can be model each other too06:00
*** pchalupa has joined #io06:26
*** republican_devil has quit IRC06:47
*** hashmal has joined #io09:25
*** springbok has quit IRC09:26
*** springbok has joined #io09:37
*** pchalupa has quit IRC10:15
*** robonerd has quit IRC10:18
*** robonerd has joined #io10:20
*** gatesphere has joined #io11:19
*** gatesphere has quit IRC11:28
*** fredreichbier has joined #io11:43
*** ijon_ has joined #io12:51
jerprologic, actually, the actor model makes no determination on the order in which things get executed, only that programs can appear sequential (i.e., block threads waiting for the result of some other computation, by way of a future or other such construct)14:15
jerprologic, it also doesn't mandate a single receiver for a single message. for instance, i implemented an actor library several years ago which operated on the principal of broadcasting a message to a group of receivers to work on it in parallel.14:16
prologicoh hi jer :)14:22
prologicyes of course14:22
prologicI don't disagree14:22
prologicgood valid points too14:22
prologicbut I was more talking about it's performance14:23
prologicand implementations I've seen14:23
prologicI've seen one in Python that uses multiprocessing and serialization to pass messages to actors running concurrently in ohter proecseses (OS threads)14:23
prologicand well it performs like a scak of bricks14:23
jerwell that's really weird, but yeah14:35
*** bjz has quit IRC14:55
*** pchalupa has joined #io15:22
*** pchalupa has quit IRC15:45
*** pchalupa has joined #io15:50
*** ijon_ has quit IRC16:08
*** asie has joined #io16:19
*** ijon_ has joined #io16:26
*** asie has quit IRC17:03
*** asie has joined #io17:03
*** asie has quit IRC17:19
*** ijon_ has quit IRC17:28
*** asie has joined #io17:34
*** fredreichbier has quit IRC18:26
*** hashmal_ has joined #io20:10
*** hashmal has quit IRC20:14
*** asie has quit IRC20:19
*** mkroehnert has joined #io20:46
prologicwell not that weird21:04
prologicserializtion will cost you21:04
prologicespecially in a higher level language like Python21:04
prologicunless you have real shared memory between processes (OS threads) as I said21:04
prologicI don't see how you'd get good performance21:04
jerprologic, no "weird" in the sense that there are better ways to transmit messages that don't involve expensive serialization21:06
jerprologic, the problem is, this generation of developers tend to be scared of things that aren't json/xml for use as an IDL21:07
jeror rather, for purposes one might also consider an IDL21:07
jerand these tend also to be the kinds of people who think message framing is hard, and sockets are hard to deal with21:08
jeronce you can get over that (none of this is a hard problem actually), then you can get some real work done21:08
prologicI happen to agree21:10
prologicI also happen to use json a lot21:10
prologicbut that's besides the point21:10
prologicI'm also aware of better ways of IPC :)21:10
prologicthere are some fantastic libraries out there for Actor Model though21:11
prologicI think there's a new one called libprocess?21:11
jeryes there are21:11
prologicwith several bindings21:11
prologiclooks good :)21:11
prologicproblem is I feel that circuits should still continue to live :)21:11
jeri've been thinking about playing a bit with taking the pi calculus and building a small language out of it, that i can use to build a new cpu21:11
prologicit's only 10 years old :)21:11
jerthat is to say, build a completely different processor which runs on top of this language21:11
prologicahh yes21:12
prologicone that optimizes lambda calculus?21:12
jernot sure what it might look like yet21:12
jerjust a passing thought i had about 8 minutes before i went to bed last night21:12
jerliterally the thought was21:13
jer"I wonder what a processor designed to mimmick the pi calculus might look like, if given a series of other computing modules it could interact with, laid out in a giant matrix"21:13
prologicjer, btw when you said "IDL" did you mean: ?21:22
jeri meant
jerthrift is a good example of an IDL21:35
robonerdjer, IDL?21:40
*** OpenSpace has quit IRC21:44
*** hashmal_ has quit IRC21:50
*** OpenSpace has joined #io21:59
*** mkroehnert1 has joined #io22:03
*** mkroehnert has quit IRC22:03
*** pchalupa has quit IRC22:08
*** c74d has quit IRC22:24
*** c74d has joined #io22:26
*** petr_ has joined #io22:28
*** cratuki has quit IRC22:35
*** cratuki has joined #io22:36
*** petr has quit IRC22:38
*** cratuki_ has joined #io22:50
*** zephyrtronium has quit IRC22:53
*** brixen has quit IRC22:53
*** petr_ has quit IRC22:54
*** petr has joined #io22:54
*** brixen has joined #io22:54
*** cratuki has quit IRC22:54
*** robonerd has quit IRC23:06
*** robonerd has joined #io23:08
*** zephyrtronium has joined #io23:09
*** mkroehnert1 has quit IRC23:33

Generated by 2.11.0 by Marius Gedminas - find it at!