On Wed, Jan 24, 2018 at 3:00 AM, Ben Goertzel <[email protected]> wrote:
> Hmm, interesting > > So if I understand right, this is just sort of like letting the PM run > independently on N processors, exploring different parts of the > hypergraph... > > Not as fancy as a parallel backtracker, but might work almost as well > in most common cases... I'd need to think/study more ... > In my experience, the greatest amount of performance gain and efficiency is gained not from micro-parallelizing small things, but running big things in parallel. Why? 1) creating things like fine-grained backtracking takes a lot of engineering effort - especially measuring the system to see what it is actually doing. If you have that kind of talent, there are more important problems it can be applied to. 2) the actual gains tend to be small, because of Amdahl's law. 3) thread-management has a non-zero and significant amount of overhead. It can swamp your problem, if the problem is small enough. And finally, the king: 4) Often, a fundamental system redesign can offer far greater performance advantages as compared to parallelizing or tuning the current system. However, this is "it depends" and is often too nebulous to do anything about. For the pattern matcher, right now, my best and brightest idea is here: https://github.com/opencog/atomspace/issues/1502 It describes an "AtomSpaceLink" that describes the membership of an atom in a particular atomspace. Such membership could be (should be) used to limit the search space, and to generally group atoms into "sets". Well, I guess this has little to do with parallelizing, other than that it has the power to control the size of the "problem space" or the "scope", thus providing a different way of controlling the performance of the system (i.e. by limiting scope). --linas > > > > > On Wed, Jan 24, 2018 at 4:46 PM, 'Nil Geisweiller' via opencog > <[email protected]> wrote: > > Regarding multi-threaded pattern matcher, I've always wondered, why not > just > > wrap the following loops > > > > https://github.com/opencog/atomspace/blob/master/opencog/ > query/InitiateSearchCB.cc#L393 > > https://github.com/opencog/atomspace/blob/master/opencog/ > query/InitiateSearchCB.cc#L634 > > https://github.com/opencog/atomspace/blob/master/opencog/ > query/InitiateSearchCB.cc#L821 > > > > in omp algos (see > > https://github.com/opencog/cogutil/blob/master/opencog/util/oc_omp.h) > since > > most of the time (always?) `found` is set to false anyway. > > > > By setting min_n appropriately (see > > https://github.com/opencog/cogutil/blob/master/opencog/util/oc_omp.h#L64 > ) > > we'd avoid unnecessary multi-threading overhead. > > > > All we'd need to have this working would be to make sure that the PM > code is > > multi-threaded safe (maybe easier said than done). > > > > What do you guys think? > > > > Nil > > > > On 01/24/2018 08:06 AM, Parth Thakkar wrote: > >> > >> I was going through the suggested projects page, and found this page: > >> (https://wiki.opencog.org/w/Parallelizing_OpenCog). I'd like to know > if it > >> is still in scope. > >> > >> -- > >> You received this message because you are subscribed to the Google > Groups > >> "opencog" group. > >> To unsubscribe from this group and stop receiving emails from it, send > an > >> email to [email protected] > >> <mailto:[email protected]>. > >> To post to this group, send email to [email protected] > >> <mailto:[email protected]>. > >> Visit this group at https://groups.google.com/group/opencog. > >> To view this discussion on the web visit > >> https://groups.google.com/d/msgid/opencog/4feba3cc-72bd- > 42fc-ab13-08fef4d81b8d%40googlegroups.com > >> <https://groups.google.com/d/msgid/opencog/4feba3cc-72bd- > 42fc-ab13-08fef4d81b8d%40googlegroups.com?utm_medium= > email&utm_source=footer>. > >> For more options, visit https://groups.google.com/d/optout. > > > > > > -- > > You received this message because you are subscribed to the Google Groups > > "opencog" group. > > To unsubscribe from this group and stop receiving emails from it, send an > > email to [email protected]. > > To post to this group, send email to [email protected]. > > Visit this group at https://groups.google.com/group/opencog. > > To view this discussion on the web visit > > https://groups.google.com/d/msgid/opencog/c143005d-d924- > daec-5961-359bdd03b253%40gmail.com. > > > > For more options, visit https://groups.google.com/d/optout. > > > > -- > Ben Goertzel, PhD > http://goertzel.org > > "In the province of the mind, what one believes to be true is true or > becomes true, within certain limits to be found experientially and > experimentally. These limits are further beliefs to be transcended. In > the mind, there are no limits.... In the province of connected minds, > what the network believes to be true, either is true or becomes true > within certain limits to be found experientially and experimentally. > These limits are further beliefs to be transcended. In the network's > mind there are no limits." -- John Lilly > > -- > You received this message because you are subscribed to the Google Groups > "opencog" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To post to this group, send email to [email protected]. > Visit this group at https://groups.google.com/group/opencog. > To view this discussion on the web visit https://groups.google.com/d/ > msgid/opencog/CACYTDBfU%2BZb76QQv60d%3D2GOg%2BVcZePC7Bwh8Y7z11oayQCxbbg% > 40mail.gmail.com. > For more options, visit https://groups.google.com/d/optout. > -- cassette tapes - analog TV - film cameras - you -- You received this message because you are subscribed to the Google Groups "opencog" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at https://groups.google.com/group/opencog. To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CAHrUA36sG6BngxN0vEEpKJA%2BnBBcaHfbhJsDc6Qi6cmsQ3KcVA%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.
