Show Idle (>14 d.) Chans

← 2018-02-15 | 2018-02-17 →
shinohai: gm trilema
BingoBoingo: Mornin' shinohai
shinohai: Buenas dias BingoBoingo .... Qntra incoming in about an hour.
BingoBoingo: Bueno, gracias.
shinohai: Plox feel free to edit as I have not advanced beyond first cup of coffee yet.
deedbot: << Qntra - CoinBase Fiat Side Glitch Opens Butts For Hurts
mod6: mornin' TSMR~
shinohai: Heya mod6 how goes?
mod6: Oh not too bad. How about you?
shinohai: Pretty good here, gonna do an offline V test in a bit once I get a few more things squared away.
mod6: Hey, cool! Let me know how it goes/have Q's.
mod6: asciilifeform: I did check if that the perf test was still running -- it was this am. I think it should finish sometime today.
asciilifeform: mod6: neat
deedbot: << Qntra - Former FBI Director Prepares Constitutional Crisis In United States With Indictments
deedbot: << Trilema - Mr. Dilworthy falls in some bad company.
mod6: evenin'
ben_vulpes: heya mod6
mod6: how goes!
mod6 is working on wrapping up all the things for V 99993
mod6 also lulz @ HUSEX
mircea_popescu: why hello #trilema!
shinohai: o hai mircea_popescu !
mircea_popescu: how goes ?
shinohai: Buenas noches y feliz Viernes
shinohai: Good here, and you?
mircea_popescu: not bad!
mircea_popescu: diana_coman "took me twice as long" ?
mircea_popescu: BingoBoingo 404s, should prolly just redirect.
mod6: mircea_popescu!
mod6: just sayin 'hi
mircea_popescu: hm mod6 did i miss a payment or what happened here ?
ben_vulpes: mircea_popescu: nah, let me take care of that. 0.8 it was?
ben_vulpes: !!pay mircea_popescu 0.8
mircea_popescu: dun dun dun!
mircea_popescu: so BingoBoingo finally got his bank acct huh !
ben_vulpes: !!v FE62C1A6B3D703CEB2F058D90938A452801E95F7DCBC03825D763D80E3F7C9DA
deedbot: ben_vulpes paid mircea_popescu 0.8
mircea_popescu: ty ben_vulpes
ben_vulpes: aha ty
ben_vulpes bbl, $child has announced bedtime by angrily tearing apart a couch to build a fort
shinohai: The plot thickens on Conbase drama:
shinohai: "We were TOTALLY not responsible at all, keep sending us monies."
mod6: thanks ben_vulpes
ben_vulpes: !!pay mod6 0.36578854
ben_vulpes: !!v EC08ABA86FC893FD5A6FCC6FB66965D075BC3B7B00C8BBE155266D7DF4D34286
deedbot: ben_vulpes paid mod6 0.36578854
mod6: tyvm!
ben_vulpes: mmmmhm
mod6: this deedbot payment system is slick asf
asciilifeform: oh hey it's mircea_popescu
asciilifeform: << this is veeeery interesting and i'd like to know what iron
a111: Logged on 2018-02-17 03:00 mod6: asciilifeform: ok it's done!
asciilifeform: mod6: and it would stand to be repeated with the files swapped ( to remove effect of, e.g., just where the given sample lives on disk physically )
asciilifeform: mod6: fwiw i have found no consistent diff, to the 6th ( beyond that, the timer is useless ) on opteron
asciilifeform: *6th decimal
mod6: sure
mod6: how do you mean 'files swapped' ?
mod6: then hypothetically the timing values for ones and zeros should simply be reversed, no?
asciilifeform: mod6: well ideally somehow to swap the inodes. still thinking how one might arrange such a thing.
mod6: iron is intel core-i5
mod6: ooh.
asciilifeform: mod6: start however by moving the two for-loops around in your tester
mod6: ok.
asciilifeform: this prolly calls for a proper test with rdtsc. but i'm saving this for 'dessert'
asciilifeform: ( after errything else, that is, is done to death )
mod6: sure. i may want to re-run this entire thing anyway... i kinda dumbly ran this on my laptop.
asciilifeform: ideally would want also not merely 'all 0' and 'all 1' but rng, and then to plot time vs hammingweight, and look for correlation
mod6: i have this mule box that i strap workloads to (core2duo box) -- should be fine as long as CPU isn't saturated, and during this test, the lappy was only like at .7 load or w/e
asciilifeform: well if it's a multicore, naturally won't saturate, proggy is singlethreaded
mod6: oh right.
mod6: anyway, on the mule, i never did get adacore running properly. this lappy is the only box i got adacore '16 running on thus far.
asciilifeform: it's ok to build on one machine and run -- on other. static .
mod6: only problem is: results aren't apples to apples.
mod6: different hardware.
asciilifeform: this is also ok. just be sure to write down, which iron.
mod6: so would want baseline + deltas all on same env.
asciilifeform: mod6: i will share my current hypothesis : all current intels have MUL leakage
mod6: aha, was thinking that's what you were thinking.
asciilifeform: otoh i could also be observing noise.
mod6: statistical noise, might be.
mircea_popescu: what is the expected gain here ?
asciilifeform: but yes this calls for a proper test, with rdtsc, then buncha MUL, then rdtsc again
asciilifeform: mircea_popescu: to find which chips one is condemned to use egyptian innerloop MUL-substitute on.
asciilifeform: currently the 2 known inhabitants of this list are ppc and arm7
mircea_popescu: how much is that worth ?
asciilifeform: well it's worth the diff b/w actual constanttime and... not
asciilifeform: i.e. leaking key.
mircea_popescu: if you find now as opposed to whenever someone runs it, this differential is worth the difference between what and what ?
asciilifeform: it is not in fact essential to find the answer ~now~.
asciilifeform: i'm putting in both variants of mul, selectably.
mircea_popescu: so what's this all about ?
asciilifeform: mod6 tried his hand at 'tester of whether ffa really does constanttime modexp'; asciilifeform comments on 1 set of output. is all.
asciilifeform: not earth-shaking experiment, no.
mircea_popescu: so his spending his time making less than minimum wage for it is basically his choice, you're just talking into the internet box ? is that the logic ?
asciilifeform: sorta like asciilifeform's spending time making no sort of wage, writing the proggy to begin with , neh
mircea_popescu: kinda, yeh.
asciilifeform: i do not order mod6 , 'go and make such a test', no.
asciilifeform: nor he -- i, 'go and...'
mircea_popescu: what's that do ?
mod6: yeah, no. it's fine. i was happy to build/run. to me this is all very exciting/important stuff. back when alf was doing the karatsuba and other things in the late summer/fall, this type of thing started to be contimplated for innerloop substitute.
mircea_popescu: mod6 you are talking to a man who spent his time dicking about and as a result is getting delorded.
mod6: i think we may need some good data collected to come up with a list of arch's that leak
mircea_popescu: if you copy his style of "work", you'll be doing yourself a significant disservice.
mod6: mircea_popescu: well, never the less, I'm still very much behind the utility of ffa
mircea_popescu: yes, but meanwhile you're engaged with bbisp, which costs $$$ each month.
asciilifeform: mircea_popescu: negrate, whatchawaiting for ?
mircea_popescu: asciilifeform separate items.
mod6: mircea_popescu: well, sure. I have a lot of things going on tho lol.
mircea_popescu: you're just inept, not dishonest.
mod6: I get your meaning though.
mircea_popescu: mod6 yes but if you don't sort them by time-productivity you'll end up in the soup.
asciilifeform: it's entirely fine afaik to negrate honest people
asciilifeform: e.g. -10 i hate his face
mircea_popescu: not how i go about it.
asciilifeform: oh while mircea_popescu is tuned in: bisp ( of which asciilifeform is 50% in ) awaits instructions re disposition of the 3 mircea_popescu boxen
mircea_popescu: iirc a) they were nsa boxes and b) the problem there was transportation, after you managed to get them locked out of the phf flight that ended up not happening
asciilifeform: mircea_popescu: transport has been arranged. what remains is 1) mircea_popescu wants'em in ? if so, how many (1,2,3?) 2) what type of disks he wants in #2 and #3 ( also remains option to outfit with disks at later time, after installation )
mircea_popescu: that doesn't say much. what exactly has been arranged, where ?
asciilifeform: ben_vulpes is making a delivery run :
a111: Logged on 2018-02-14 17:48 ben_vulpes: i have one 1U slot open in my next courier run to .uy, if anyone wants their machine couriered down speak NOW or i'm going to fill it with a box of my choice
asciilifeform: i will let him speak re exact time.
mircea_popescu: so how many of 1, 2 or 3 do i want to fit in a 1u slot ?
asciilifeform: well if 1, it's a 1u subscription, if 2 -- 2u, so forth
mircea_popescu: anyway. it's unclear to me these are even dc worthy in any sense.
asciilifeform: if mircea_popescu wants them at his house to use as boat anchor -- oughta specify this. they dun do much good here on desk.
mircea_popescu: much like errything else.
ben_vulpes: what's so not-dc-worthy about 'em?
asciilifeform: ben_vulpes: apparently , that asciilifeform touched'em
asciilifeform: mircea_popescu is apparently determined to maliciously run s.nsa into the ground, and pin the corpse on asciilifeform .
asciilifeform: and yes, asciilifeform -- not quite ~yet~ stripped of his stars and epaulettes -- wishes to put this matter before the forum.
asciilifeform: supposing we are still carrying on with a pretense of a forum, and a lordship.
mircea_popescu: anyway. the situation is that s.nsa naively bought some iron at your recommendation at some prior point. meanwhile you refused to deliver the "here's the mega-article comparing all hardware, which will guide future republican hardware building.
mircea_popescu: and seeing how you're the fellow who asks me how many items do i want to fit in ben_vulpes 's one item slot, i have serious doubts about your cogency anyway.
mircea_popescu: so for now, waiting to see what bbisp decides to do re hardware, and vageuly considering getting out-of-house expertise.
mircea_popescu: ben_vulpes i have nfi. could take too much power, could whatever the fuck.
asciilifeform: ben_vulpes in point of fact allocated 2 slots for s.nsa . and was asking re whether 3rd ought to also go to it, or to another.
mircea_popescu: we still don't have a proper item on which to base hardware strategy.
mircea_popescu: and i wouldn't trust asciilifeform to write it now, in any case.
asciilifeform: and the measured draw is 150W . ( as descibed in detail in the past. )
asciilifeform: mircea_popescu you are not a party to bisp.
mircea_popescu: i still intend to follow their standards.
asciilifeform: a customer (possibly ! not yet subscribed) -- yes
mircea_popescu: what's "being a party" has to do with anything ?
asciilifeform: asciilifeform sets the iron standards.
asciilifeform: aaaactually.
mircea_popescu: in no case do you set any standards.
mircea_popescu: that train left.
asciilifeform: so mircea_popescu , having sold his stake, and not even yet a subscriber, wishes to decree re what is done at bisp.
mircea_popescu: i guess, perhaps, the standard for time frittering. but that's well set already, so.
mircea_popescu: speaking of which, dja have some sort of decision yet ben_vulpes / mod6 ? are you goint to offer box rentals ? what kind ?
asciilifeform: it so happens that asciilifeform has purchased a 50% stake in bisp. and specifically with one condition: that mircea_popescu (named, concretely) will not be a back-seat driver there.
mod6: mircea_popescu: that is being discussed, yes.
mircea_popescu: well, that was pretty dumb on their part, i guess, but whatever, experience is apparently difficult to process.
ben_vulpes: will be renting boxes out, this first courier run is filling up with space-rental.
mircea_popescu: ben_vulpes ok, but which are you buying ? or what's teh strategy exactly ?
ben_vulpes: i want the next run to have one box with lots of ram and cpu to start selling vps and shared hosting in earnest, and another 2 of smaller boxes that can be rented out entire in the $3-400/mo range.
mircea_popescu: yes, but the problem with "lots of x" is that...
mircea_popescu: so basically is the idea that you'll just do whatever and we're missing this opportunity to attempt an' standardize iron ?
← 2018-02-15 | 2018-02-17 →