#trilema | 2019-02-08

01:18

http://bimbo.club/2019/02/the-travel-guide/ << Bimbo.Club -- The travel guide

~ 1 hours 17 minutes ~

02:35

asciilifeform, one could certainly keep at all times 64 worker threads idling and waiting for jobs; initially I had implemented it precisely as such: all it takes is to accept a Get_Job request only when there IS a job to hand out; it adds a bit of complexity in there though because there has to be a flag for "available job" and then it has to be kept up to date correctly (which can be a bit tricky as it's not only when a job is added, it'

02:35

diana_coman

s also when a job is finished since the player id may be released or when/if a worker aborts unexpectedly )

02:36

diana_coman

re workers aborting unexpectedly: it shouldn't happen, no; but I can't rely on it not happening

02:39

diana_coman

that being said, I'm not sure in fact that there is a need for the Initialize procedure itself if I'm to cut something out (it's enough to initialize the array when declaring it - although technically it will happen later than within initialize,it's still before anyone gets to use it so perfectly adequate)

02:40

01:18	feedbot	http://bimbo.club/2019/02/the-travel-guide/ << Bimbo.Club -- The travel guide
		~ 1 hours 17 minutes ~
02:35	diana_coman	asciilifeform, one could certainly keep at all times 64 worker threads idling and waiting for jobs; initially I had implemented it precisely as such: all it takes is to accept a Get_Job request only when there IS a job to hand out; it adds a bit of complexity in there though because there has to be a flag for "available job" and then it has to be kept up to date correctly (which can be a bit tricky as it's not only when a job is added, it'
02:35	diana_coman	s also when a job is finished since the player id may be released or when/if a worker aborts unexpectedly )
02:36	diana_coman	re workers aborting unexpectedly: it shouldn't happen, no; but I can't rely on it not happening
02:39	diana_coman	that being said, I'm not sure in fact that there is a need for the Initialize procedure itself if I'm to cut something out (it's enough to initialize the array when declaring it - although technically it will happen later than within initialize,it's still before anyone gets to use it so perfectly adequate)
02:40	mircea_popescu	the problem is, since you can't guarantee workers staying up, you'll need a way to create them anyway.
02:40	mircea_popescu	it is actually better to have a robust life.death cycle than to pretend like death ain't supposed to happen.
02:40	diana_coman	re workers and specialisation: the worker tasks are not specialised, they do whatever job they are given and moreover the choice is made by job type as it were
02:40	diana_coman	aha
02:40	mircea_popescu	there's nothing to force you to kill them any more often than you benefit from anyway.
02:41	diana_coman	in mircea_popescu's model/apache terminology the "workers" are in fact my jobs there: they get created , executed, finish
02:42	diana_coman	I suspect asciilifeform's point was more along the lines: why kill them at all instead of letting them idle in there until something to do; not really "why do you bother having a way to create them?"
02:42	mircea_popescu	what ? no, not at all.
02:43	mircea_popescu	apache workers ain't jobs ; and i took alf's comments to mean "why are you killing them 9k times/s", which is in his usual style of random assumptive approach, nobody said they'd be killed often.
02:44	diana_coman	if there's nothing for them to do, they get killed; but not sure what's the point in keeping them idling anyway; i.e. if it's busy then no, they don't get killed
02:45	mircea_popescu	right.
02:45	diana_coman	I'll probably need to find out exactly what the overhead for their creation is anyway
		↖
02:46	mircea_popescu	i suspect you'll end up with them specialized, rather than case-switched, on acct of the overhead involved. however, time wil ltell
02:46	diana_coman	overhead of the case-switched/
02:46	diana_coman	?
02:46	diana_coman	from the job market's pov it's easier to have them not-specialised really
02:47	mircea_popescu	yes. because ada entities keep getting wrapped in things that end up eating runtime to "elaborate" and whatever ion thios vein.
		↖
02:47	mircea_popescu	that much is true.
02:47	diana_coman	and the case-switch is done inside the worker so on their time at most
02:47	diana_coman	hm, you mean the overhead at creating the job itself then (as that's the object that gets bigger to elaborate)
02:47	mircea_popescu	every time you load one you have to load the whole pile of everything, for the whole case set. eventually (it stands to reason) you'll find it cheaper to specialize them.
		↖
02:47	mircea_popescu	but really, no rush with this bikeshed.
02:47	diana_coman	aha
		~ 6 hours 20 minutes ~
09:08	auctionbot	Buy order # 1037 has ENDED: No sale. Attn: BingoBoingo
09:12	BingoBoingo	!Xbuy 308mn 24 1k Wired Filthy Fiats (WU esta bien)
09:12	auctionbot	Buy order # 1038 created by BingoBoingo: 1k Wired Filthy Fiats (WU esta bien) Opening: 308mn ecu Ending: 2019-02-09 02:16:24.769899 UTC (23 hours)
09:12	BingoBoingo	^ Auction take 2
09:12	diana_coman	http://btcbase.org/log/2019-02-07#1893544 -> adding to this, on (re)^n -read of docs I doubt there's any benefit in having a controlled type in there; essentially the main thread will wait for all tasks spawned by the SJM to finish from what I understand; so then aborting them in a Finalize makes precisely 0 sense and instead I should probably make it a plain array and offer users of the package a non-synchronised procedure to abort them
09:12	a111	Logged on 2019-02-07 23:56 asciilifeform: http://ossasepia.com/2019/02/07/seppuku-job-market-minimal-dynamic-tasking-in-ada/#selection-109.1672-109.1738 << will be interesting to see if this functions if stuffed into a static lib, in light of the http://btcbase.org/log/2019-01-22#1889467 puzzler
09:12	diana_coman	if /when desired (i.e. at shutdown time, the sequence might be something along the lines send stop signal to sjm and then /after a while, if there is anything still running, call the abort)