#lisp - Thu 12 Apr 2007 between 17:06 and 17:15



Fareand currently using ad-hoc functions to generate what task to assign the next available worker
michaelwFare: are you asking about load balancing?
rahulmichaelw: no, fully distributed computing
erlang-esque
Dazhbogoo
michaelwrahul: um
Faremichaelw, yes
rahulI guess you could _call_ it load-balancing, but it's not the same as an http load balancer unless you think of the web servers requesting the farm for sub-requests
pisofroydnj: in tiger.lisp, you've got some calls to LIST that take more arguments than CALL-ARGUMENTS-LIMIT is required to accept
michaelwrahul: i did not say http at all. load balancing is a general concept.
rahulmichaelw: right, http as an example request protocol
pisofroydnj: could you replace (list ...) with '(... )
?
(or do you modify those lists?)
rahulmichaelw: most load balancers don't shift subtasks around the farm
Fareso, cells and compute-class.
Thanks
rahulan entire request is handled by a single physical server
xirieFare: http://common-lisp.net/project/computed-class/
froydnjpiso: in the MAKE-ARRAY calls for the sboxes?
michaelwrahul: please point me to some of those load balancers
rahulmichaelw: which ones?
michaelwFare: googling for load balancing and/or control theory might point you in the right direction. usually some prediction of the amount of work to be expected is needed
rahul: of the "most" you mentioned
cmmis computed-class a fork of cells or something?
attila_lendvaixirie: did you use computed-class?
rahulmichaelw: the F5 http load balancer is the 'typical' one
piso(defparameter tiger-t4 (make-array ... :initial-contents (list ...)))
attila_lendvaicmm: it's written from scratch
froydnjpiso: ok. yes, those could be changed to quoted lists
xirieattila_lendvai: not yet (have a project due next week), but it looks a lot cleaner than cells.
Faremichaelw: no, I don't need any of that
cmmfork all the way down! :)
rahulFare: actually, cells only knows what dependencies there are after you compute the result
Fare: it only uses dependencies to know when it needs to update
pisofroydnj: xcl's CALL-ARGUMENTS-LIMIT is 256
(although maybe there's no good reason for that)
Farewhat I need is a simple framework to build "nodes" that depend on other "nodes" for completion, sometimes in parallel, sometimes in series, sometimes with more complex dependencies
with new nodes appearing on a need basis, and disappearing when completed
michaelwFare: a dependency graph and breadth/best-first traversal then? :)

Page: 5 12 19 26 33 

IrcArchive