home bbs files messages ]

Forums before death by AOL, social media and spammers... "We can't have nice things"

   comp.ai      Awaiting the gospel from Sarah Connor      1,954 messages   

[   << oldest   |   < older   |   list   |   newer >   |   newest >>   ]

   Message 946 of 1,954   
   russell kym horsell to David   
   Re: balancing number of Agents   
   05 Mar 06 06:29:20   
   
   XPost: sci.cognitive   
   From: kym@ukato.freeshell.org   
      
   In comp.ai David  wrote:   
   ...   
   > As far as theoretical approaches to "balancing N" go, you could look   
   > at Queueing Theory.  Given probability distributions over buyer   
   > arrival times and service durations, it allows you to predict buyer   
      
   "Predict" in only in the most trivial cases, of course. ;)   
      
   > waiting times if N sellers are available.  That should allow you to   
   > set the overall parameters appropriately, which you can improve by   
   > the kind of adaptive approach outlined above.   
      
      
   An "AI solution" could again be adopted as far as the OP and O2P are   
   concerned.  E.g. client response time could be used to adapt a   
   "startup" and "shutdown" agent/expert. As in most of this kind of AI,   
   beware corollaries of Arrow.   
      
      
   Another idea I've used myself is just a simple slant on IPD theory.   
   The servers are shared resources. Problems with competing clients   
   sharing resources can (and typically ;-) fall under the criteria for   
   a prisoners dilemma. If 2 clients both ask for the cake now, they   
   may both suffer a penalty. If one asks and the other doesn't, the one   
   not asking gets a bigger penalty and the "winner" gets the whole cake.   
   If the agents agree to take turns they both -- in the long run -- do better   
   than otherwise.   
      
      
   I've used small finite automata to share a fixed number of resources in IPD   
   situations. Like comms channels. Works as well as exponential backoff in most   
   cases, and better in certain cases.   
      
      
   In the current context, part of the penalty/payoff (perhaps an amortised cost   
   if   
   servers speculatively hang around in process trees) is startup time for   
   a new server that removes an otherwise access conflict. Should be just a   
   minor adjustment well worthy of a PhD or 2 and slew of 2nd-tier papers. ;-)   
      
   [ comp.ai is moderated.  To submit, just post and be patient, or if ]   
   [ that fails mail your article to , and ]   
   [ ask your news administrator to fix the problems with your system. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)   

[   << oldest   |   < older   |   list   |   newer >   |   newest >>   ]


(c) 1994,  bbs@darkrealms.ca