log☇︎
4700+ entries in 0.002s
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934772 << neato, trinque . plz consider to genesis the thing, so can have multiple redundant pastetrons, as with logger
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934699 << in the orig. '17 thread re 'make cpu', mircea_popescu made the observation 'the sticker prices have 3 or 4 extra zeros on'em', i.e. pointless to even ask the q before we have penetration into the asian snakepits where iron is actually baked. without such, we're stuck looking at the chump ratecards.
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934692 << the tricky bit is that to sell sumthing, it gotta exist 1st. ( granted various hucksters manage to 'sell' thin air, but this is outside the scope imho )
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934677 << this is entirely troo ( recall the infamous ibm derp's 'there is a world market for perhaps 6 computers' ? )
asciilifeform: the orig q was re line rate of nic. but even the modest piz pipe , 200Mbit/s , is good for ~50k 512B packets / sec , when otherwise unloaded.
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934686 << yer figure is unrealistically pessimistic, 1k * 512b == 512kB/s . there indeed is overhead in nic etc, but udp packets can be received at very close to rated capacity of pipe.
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934668 << this is serious bugola. i'ma take a look.
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934664 << i also haven't learned hieroglyphs and moved to pekin. doesn't mean that couldn't, or at some point won't come to it.
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934661 >> e.g. http://www.greenarraychips.com/ . or the fact that his book was translated and printed in run of 1e6 even in sovok. but i expect 'not famous enuff' to mircea_popescu , cuz aint in his personal kit ? or wat.
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934655 << appeal of forth is that there ~aint~ a 'tower'. i.e. can stand up all the compiler you need in coupla 100 bytes.
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934654 << if you dun see sumthing with naked eye, dun mean it aint there.
asciilifeform: doesn't figure prominently in opensores liquishit -- is tru
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934647 << thing has install base that dwarfs apache. ( it's in the bootloader of erry single sgi, sun, even crapple box sold to date; in 1e6 pieces of embedded equip., and whole buncha 'unsexy' irons, from mains transformers, burglar alarms, etc., possibly even in yer mercedes. )
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934643 << forth will be ~5~ decades next yr.
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934642 << if it's to live, gotta cement it into trinque's collection, yes
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934617 << i dun have a serious output for that q yet, will come back to it .
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934613 << why ty.
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934607 << interesting idea, but plox to give detailed spec, i suspect no one can bake based on the 3ln given
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934537 << this is troo, could cache search pgs at the current rate and not fill disk in 100y
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934490 << also relevantly, if ben is down w/out being rebootable, mod6 needs a new co-chair in tbf
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934477 << archaeological find. manufacturer was 1 of the 2 castles where sovok built nukes , accelerators, related hardware.
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934474 << but, lol, actual mircea_popescu never says anyffin of the kind -- always gives concrete barf when finds bug. so why then.
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934457 << in orig q i was thinking 'does this personage correspond to a wot key', prolly oughta have said concretely. ( if mircea_popescu had revealed the chix's human name, this would not have meant anyffin to asciilifeform et al )
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934436 << the american generals of '50s had somewhat similar concept, 'ooda loop'
asciilifeform wakes up, eats backbreaker l0gz..
asciilifeform will comment on mircea_popescu's other observations when wakes up.
asciilifeform: http://logs.nosuchlabs.com/log/trilema/2019-09-07#1934468 << ha! fewer things to do! i aint about to protest.
asciilifeform: ( fleanode apparently surrounds 'action' msgs with 2 spurious 0x01, one on ea. side )
asciilifeform: iirc phf suggested a fix, but i have not tested yet
asciilifeform: mircea_popescu: ctcpism
asciilifeform off to meat.
asciilifeform: ( pet : 'when me?' )
asciilifeform: trinque: i'ma also bbl. food, sleep...
asciilifeform: see also
asciilifeform: if they'd actually made it variant, like was orig. meant, you would have noticed right away that your algo aint it.
asciilifeform: it is interesting, irc protocol asks that pingism follow form 'PING foo' 'PONG foo' . but their foo seems to be a constant.
asciilifeform: ( in spite of algo being wrong )
asciilifeform: trinque: if you had made it 29 -- prolly would have seemed to work 4evah
asciilifeform: ahahaha
asciilifeform: trinque: i found , when experimented, that you have just about half minute! to answer their ping
asciilifeform: as in, single digits
asciilifeform: iirc mircea_popescu surveyed, and found that it was some laffably small #
asciilifeform: how many even work any moar
asciilifeform: would depend which fleanode box, neh
asciilifeform: ( i imagine ping from asia, ~decreased~ neh )
asciilifeform: ping to trinque's current chair ?
asciilifeform: would really like to get back to conveyor .
asciilifeform: after http://logs.nosuchlabs.com/log/trilema/2019-09-06#1934226 is put to rest, i'ma leave the logger alone, unless someone finds catastrophic problem .
asciilifeform: seems like erry summer asciilifeform does a jet-powered flight into brick wall. last time it was the cr50 thing.
asciilifeform: trinque: it is. tho i lost more than month (not counting the entirely pointless expedition of 'M', then could say lost 3mo ! )
asciilifeform: i.e. even answering the feasibility q is likely to be quite expensive.
asciilifeform: whole thread nearly borders on sin, given as almost anyffin i say on the subj is likely to be catastrophically wrong.
asciilifeform: 1 of the reasons why iron is such a bitch is that ~very~ small differences in logical design can give gigantic diff in performance.
asciilifeform: this is difficult to calculate 'on napkin' w/out an actual complete iron design, however.
asciilifeform: entirely possib. that it is only feasible in 35nm or better.
asciilifeform: trinque: opposite.
asciilifeform: aaand this is counting whole interval, in which not only rsa but e.g. keccak padding calc. gotta happen.
asciilifeform: ( btw , for thread-completeness -- you need 4096 modexp in < 3.8 ~microsecond~ to eat GB at line-rate. )
asciilifeform: i certainly wouldn't invite anyone to pay to build such space station until thoroughly familiar with working example 'in soft'
asciilifeform: a gb nic can theoretically swallow 2**18 512b packets / sec .
asciilifeform: i'ma get it to the point where does the job orig. promised. but 'unsinkable network battleship' will have to wait for actual irons , as i understand .
asciilifeform: lol
asciilifeform: you can parallelize the work, but the amt of effective diff it makes , is relatively small when compared to the absolute size of the above shortfall.
asciilifeform: but important to remember that yer nic can still eat 512b shitograms roughly 10,000x faster than you can modexp on'em . even w/ bvt's asmisms.
asciilifeform: trinque: the slow prototype is what asciilifeform (is slowly getting back nao!) to working on .
asciilifeform: on ordinary x86 -- state-of-art today is bvt's inline asm where 0.25s per 4096bit modexp .
asciilifeform: if you actually want to verify 4096bit rsa packets at line rate, tho, yer stuck baking silicon.
asciilifeform: imho it's the obv. Right Thing.
asciilifeform: trinque: it would indeed 'cut world', in the sense that gives you a gossiptronic net fully layered over the traditional one. and with dozen+ entrance boxen, suddenly impossible to effectively ddos anything on the inner (i.e. what comes out of the 2nd plug) world.
asciilifeform: *the last one
asciilifeform: i'm inclined to agree. but possibly asciilifeform is that last one who oughta talk about 'what has market', asciilifeform also thought FG had mega-market waiting.
asciilifeform: i proposed it in '16 as a 'gateway' into gossipnet. ( eats packets at line rate, emits from 2nd nic hole the ones that were found edible. )
asciilifeform: 'iron gossipd'.
asciilifeform: iirc we had thread about this.
asciilifeform: their physical size is ~cube of the bitness , and speed of light is finite.
asciilifeform: aha. tho presently i doubt that it's physically possible to make a 4096x4096b multiplier converge in <1ns
asciilifeform: this figure for e.g. 65nm . i.e. ~gz or so, realistically.
asciilifeform: trinque: so, actual honest iron ? 1-2mil transistors. i.e. coupla 100k $ for 1st tray, and afterwards a few cents ea. if baked in qty. as you might expect.
asciilifeform: ( http://logs.nosuchlabs.com/log/trilema/2019-09-06#1934333 is obv. exaggeration, if anyone doesn't grasp. but yer still talking about a 1ms op where normally x86 gives 1ns )
asciilifeform: i can maybe answer the q, if it is made very, very concrete.
asciilifeform: trinque: elaborate re 'as prosthetic' ?
asciilifeform: but on x86 ?
asciilifeform: trinque: an 'iron peh', where the ALU is actually e.g. 4096b-wide -- may be workable as general-purpose comp.
asciilifeform: tho imho in this case it is moar like using battalion howitzer as fishing pole.
asciilifeform: iirc mircea_popescu called this 'use 50cal rifle as fishing pole'
asciilifeform: ( or even 0x0 )
asciilifeform: and wait ~1s to multiply 1x1 ?
asciilifeform: rly you'd want to burn e.g. 4096bits when looping over a 1byte variable ?
asciilifeform: ugh
asciilifeform: trinque: trivial if you know the magick inits for $nic.
asciilifeform: on e.g. x86, a forth complete enuff that you could comfortably live the rest of your life in it, weighs about 1kB.
asciilifeform: ( moor, if he were in his grave, would be spinning like propeller )
asciilifeform: 'peh' is -- arguably -- a (very fascist) forth .
asciilifeform: so 'bootstrap problem' evaporates.
asciilifeform: forth also has the afaik unique attribute where standing up a forth on bare iron is ~weekend's work for an adept
asciilifeform: complete w/ electromagnetic models etc
asciilifeform: chuck moor had , famously , complete vlsi production plant in iirc 3k loc of forth .
asciilifeform: i'm not about to ask folx to learn it tho ( asciilifeform in fact quite fond of forth, and used extensively. but i also recognize that to most people utterly 'martian' )
asciilifeform: trinque: for overall smallest mass of system + all needed for $problem ? afaik forth in fact is the unchallenged champ, wherever was historically tried.
asciilifeform: well i was genuinely curious. for all i knew, lighter