#asciilifeform | 2021-01-04

09:34

http://thetarpit.org/2021/re-ivanovna-et-al-2020 << The Tar Pit -- Re: Ivanovna et al., 2020 (DOI: 10.7554/eLife.58906)

~ 1 hours 15 minutes ~

10:50

http://logs.nosuchlabs.com/log/asciilifeform/2021-01-03#1028160 << mno, not by 'one access for data and code.' but by fact that a) cpu (in abstract) can do 1 thing at a time b) attempts to magick away (a) by adding multiple cpu, face contention for shared resources (memory; i/o) .

10:50

snsabot

Logged on 2021-01-03 22:16:22 verisimilitude: Say, asciilifeform, the von Neumann bottleneck is caused by the one access for data and code; it doesn't feel meaningfully different to write programs which don't use memory for anything beyond code, however.

10:51

09:34	feedbot	http://thetarpit.org/2021/re-ivanovna-et-al-2020 << The Tar Pit -- Re: Ivanovna et al., 2020 (DOI: 10.7554/eLife.58906)
		~ 1 hours 15 minutes ~
10:50	asciilifeform	http://logs.nosuchlabs.com/log/asciilifeform/2021-01-03#1028160 << mno, not by 'one access for data and code.' but by fact that a) cpu (in abstract) can do 1 thing at a time b) attempts to magick away (a) by adding multiple cpu, face contention for shared resources (memory; i/o) .
10:50	snsabot	Logged on 2021-01-03 22:16:22 verisimilitude: Say, asciilifeform, the von Neumann bottleneck is caused by the one access for data and code; it doesn't feel meaningfully different to write programs which don't use memory for anything beyond code, however.
10:51	asciilifeform	http://logs.nosuchlabs.com/log/asciilifeform/2021-01-03#1028161 << register, in the typical sense, is simply an expensive memory.
10:51	snsabot	Logged on 2021-01-03 22:16:58 verisimilitude: That is, be there anything much special about that class of programs which, after initialization, have all mutable state live in registers and whatnot?
		~ 59 minutes ~
11:50	asciilifeform	verisimilitude: i'll add what the schoolbook ~won't~ tell you : the shared-resource bottleneck is ~not~ the greatest catastrophe of the vn model of computation. rather, it is the shit-sandwich of broken abstractions based on the interrupt.
		↖ ↖
11:52	asciilifeform	interrupts make the very notion of a deterministic process on the machine, a virtual impossibility.
11:54	asciilifeform	1st-law compliance is difficult (and in practice, impossible) on a vn box.
11:55	asciilifeform	even pc iron designers conceded this, and offloaded all kinds of processing from cpu to special-purpose micros (in nic, disk, kbd, etc)
11:59	verisimilitude	I recall reading an article by Djikstra about how, decades ago, the machine designers weren't implementing interrupts correctly.
12:00	verisimilitude	I never thought of interrupts as particular to the von Neumann model, but I suppose this is due to the dearth of other machine types more than anything.
12:06	asciilifeform	verisimilitude: entirely particular to vn model. observe again, e.g. fg has no notion of interrupt, because has no notion of instruction flow.
12:10	asciilifeform	shinohai: bot awol ?
12:15	shinohai	ty asciilifeform
12:15	asciilifeform	verisimilitude: vonneumannism is not simply a set of broken mechanisms, but of broken ~thought patterns~. there are certain problems that can't even be reasoned about effectively for so long as operating with concepts where dichotomy b/w 'cpu' and 'memory'.
		↖
12:15	snsabot	(trilema) 2015-02-22 asciilifeform: back to thread, momentarily, before i nod off to sleep; certain problems (i dare say most of the 'big ones') in algorithms magically dissipate when you burn the retarded von neumann thing
12:16	asciilifeform	wb btcinfobot
12:16	asciilifeform	$ticker btc usd
12:16	btcinfobot	Current BTC price in USD: $30922.83
12:16	asciilifeform	cheap!11111 again
12:17	verisimilitude	Provide an example of such a problem; none currently come to my mind.
12:18	asciilifeform	verisimilitude: iirc r. feynman's example. he gave a conference table of 'computer scientists' a bundle of sticks, of various lengths, to sort by length. asked about the complexity class of the problem.
12:19	asciilifeform	verisimilitude: do you know this story ?
12:19	verisimilitude	I don't believe so, but ``spaghetti sort'' comes to mind.
12:19	asciilifeform	aha, it.
12:20	asciilifeform	sorting is in o(n). unless, of course, yer trapped in a vn machine..
12:21	verisimilitude	I'd always thought of that as a consequence of analog vs. digital, not like this.
12:22	asciilifeform	( i'll add to the story -- if yer ~not~ trapped in a vn machine, then sort not even o(n) -- if you can simply tell the sticks to 'walk forward if you see a taller stick in front of you; otherwise stand still' and they take ~at most~ N steps ! )
		↖
12:22	asciilifeform	verisimilitude: it aint.
12:22	asciilifeform	digital does not necessarily imply vn cripplement.
12:23	verisimilitude	Well, I suppose this has still educated me a tad; I appreciate it, asciilifeform.
12:23	asciilifeform	np
12:24	verisimilitude	Some examples of other architectures I've seen is giving the memory minor computation abilities, such as clearing or pattern-embedding; that could be neat.
12:24	verisimilitude	I like the idea of being able to tell the memory to set the current memory incremented from the last memory, for N steps, and so having the APL iota in hardware.
12:26	asciilifeform	verisimilitude: would be great, except that, barring miracles, 0 of these academic curios will be fabbed in anyffin like a modern process.