Logs for #trilema

Go to: #trilema #trilema-mod6 #trilema-trinque #chainstate #eulora

2017-4-25

danielpbarron: apparently one of the firms developing that tech has a building at 666 5th Ave NYC
mod6: my second run on the 2nd FG out of my bunch has completed. results here: http://www.mod6.net/fg/fg-test/fg2.ent_run2.txt http://www.mod6.net/fg/fg-test/fg2.dieharder_run2.txt
asciilifeform: neato mod6
asciilifeform: incidentally i had a thought re what an 'ideal dieharder' might look like.
asciilifeform: imho it ought to return an ~algo~ for attempting a better-than-chance prediction of nth bit, given n;
asciilifeform: or of k-th bit, given bits k_i-j ... k_i , for some j.
asciilifeform: because that's, really, what alleged 'weakness' means -- that you can say 'can predict...'
asciilifeform: this generalization also exposes what exactly the 'minimal sample' is, likewise.
asciilifeform: for very short input, algo will be likewise short. for long, well-entropic -- long. for long, poorly-entropic -- short.
asciilifeform: the scalarization of 'weakness' is an absurdity. gotta make this -- specific -- claim.
asciilifeform: * of k_i+1st bit
asciilifeform: ideally this output (algo) would consist of a runnable lisp proggy.
asciilifeform: that way you do not have to wonder wtf the particular test consisted of. and can immediately replicate.
asciilifeform: (without a copy of the tester, even.)
asciilifeform: btw i now realize that this entire idea isn't asciilifeform's, but actually greg chaitin's.
asciilifeform: from 1990s.
asciilifeform: described in almost exactly these terms.
asciilifeform: but i haven't the original paper at hand, this is a 20y+ memory.
asciilifeform: ( if anyone recalls where this was -- plz post !! )
asciilifeform: until somebody suggests a better name, i'ma call this hypothetical method 'die maximally hard'.
asciilifeform: (dmh).
mod6: heheh. i'll look for the posting
asciilifeform quite enjoyed the ru sphere's name for dieharder, 'умри тяжело'
asciilifeform: ( approx. 'go and die a hard death' )
asciilifeform: poor ol' english folx, have not only toilet and sink in same rooms, but imperative same as infinitive verb...
mircea_popescu: asciilifeform this is a pretty hard standard though.
mod6: not sure that this is it, but a decent read anyway: https://www.cs.auckland.ac.nz/~chaitin/sciamer.html
mod6: !~s Ada Big_Number
jhvh1: mod6: Error: "s" is not a valid command.
mod6: inconceivable
shinohai: !~google Ada Big_Number
jhvh1: shinohai: Arbitrary length integer in Ada - Stack Overflow: <http://stackoverflow.com/questions/1309195/arbitrary-length-integer-in-ada>; The Big Number Page: <http://bignumber.chez.com/>; Ada –A Crash Course - VTC - Vermont Technical College: <http://web.vtc.edu/users/pcc09070/TutorialAda/AdaCrash.pdf>
mod6: I guess what I was looking for was this: http://btcbase.org/log/2017-03-29#1633709
a111: Logged on 2017-03-29 12:08 asciilifeform: http://btcbase.org/log/2017-03-29#1633699 << it is a notbad library to study (no external deps!) -- but it DOES use heap.
mod6: I was thinking today about a ~tmsr~ crypto lib, and it starts (probably) thinking about bignum
mod6: I did run across one on shithub that was implemented with a linkedlist. this seems like exactly what we wouldn't want.
asciilifeform: http://btcbase.org/log/2017-04-26#1647926 << funnily enough, it's what i'm polishing for release as we speak.
a111: Logged on 2017-04-26 02:18 mod6: I was thinking today about a ~tmsr~ crypto lib, and it starts (probably) thinking about bignum
mod6: !
mod6: That's exciting asciilifeform
asciilifeform: since mod6 mentioned subj, i'll say a bit
asciilifeform: thing is optimized for -- strictly -- constant (always-worst-case) time and space usage; and fits-in-head (in that order)
asciilifeform: so not only , e.g., comparison, but addition, subtraction, take same number of ops regardless of what bits are set
asciilifeform: and all ints have fixed max width ( set at warmup )
mod6: aha. ok.
mod6: this is pretty similar to the semi-counterintuitive idea i had been thinking about -- or more probably, was remembering from a previous discussion in here about the same.
asciilifeform: no normalization, either - adding 8kb of zero to another 8kb of zero takes same time as any other possible pair of 8kb bitstrings.
mod6: right on. very cool.
asciilifeform: and no seekritbit-dependent branches.
asciilifeform: anywhere.
asciilifeform: ( in the arithm ops, that is )
asciilifeform: for some peculiar reason, everybody else (afaik) who implemented a bignumtron, only did this (or claimed, at any rate) for modular exponentiation -- but not for its subcomponent ops
asciilifeform: the usual argument is 'that'd be slow'
asciilifeform: but that's transparently bogus, rsa has 0 business ever running in anything but worst case spacetime
asciilifeform: ( and on encrypt, as well as decrypt end ! leaking plaintext isn't more forgivable on transmitter end than on receiver )
mircea_popescu: so how slow is it ?
asciilifeform: mircea_popescu: the only palpably slow knob is prime-finding
asciilifeform: (several min. on pc with no entropy bottleneck (i.e. FG present))
mod6: my re-run of eatblock is complete... will re-post stats etc here in a bit
mod6: asciilifeform: cool
asciilifeform: aim is to cap the thing at its current ~1000 ln.
asciilifeform: short and simple enough for even exceptionally impatient folx to read and grasp.
mod6: very exciting stuff
asciilifeform: actually-- deadly boring stuff. and intentional. 0 surprises. 'schoolboy' algos for just about all arithm ops.
asciilifeform: http://btcbase.org/log/2017-04-26#1647918 << was quite exactly this idea, but it also had the lisp outputs iirc
a111: Logged on 2017-04-26 01:56 mod6: not sure that this is it, but a decent read anyway: https://www.cs.auckland.ac.nz/~chaitin/sciamer.html
asciilifeform: http://btcbase.org/log/2017-04-26#1647917 << imho it's the only sane standard. rng 'weakness' can only be expressed properly as a 'predictor'. anything else is 'suggestively named token'
a111: Logged on 2017-04-26 01:54 mircea_popescu: asciilifeform this is a pretty hard standard though.
mod6: <+asciilifeform> actually-- deadly boring stuff. and intentional. 0 surprises. 'schoolboy' algos for just about all arithm ops. << that's just it though. excitingly boring.
asciilifeform: lol!
asciilifeform: mircea_popescu: so, to take example, if you found a rng that turns out to emit digits of pi -- the proof-of-weakness will be plouffe's algo.
mircea_popescu: plouffe's algo took until 2010 to be spit out. by plouffe.
mircea_popescu: notwithstanding it is trivial in that "olympiad problem" sense.
asciilifeform: 2001 iirc
mircea_popescu: right.
mircea_popescu: my 2001 is alt-spelled.
asciilifeform: but note that it dun have to be 'nth digit' , can also be 'next digit', o(n) , works just as well.
asciilifeform: for weakness proof.
mircea_popescu: the principle of the thing though. it is an exceptionally hard standard
mircea_popescu: not without theoretical merit ; but in practice imponderable.
asciilifeform: on the contrary -- all of the classical tests can be cleanly rewritten in this form
asciilifeform: at least, all of the meaningful ones...
asciilifeform: we aren't speaking of an ~optimal~ predictor, note
asciilifeform: ( that's provably undecidable )
asciilifeform: but of ~any~ better-than-chance predictor.
asciilifeform: any rng test can be rewritten as a producer of a predictor. or the test is rubbish.
asciilifeform: possibly the pi example was a poor one -- no ~automated~ tester, will ever find it, unless specifically made to
asciilifeform: ( the digits of pi do not autocorrelate, as far as anybody ever publicly found, in any other way )
ben_vulpes: https://blog.buoyant.io/2017/04/25/whats-a-service-mesh-and-why-do-i-need-one/ << "everything must run over http" monomania in full effect
ben_vulpes: parts of it have an odd feel, as though i were reading altcoin propaganda from another dimension
mod6: Ladies and Gentlemen of The Most Serene Republic: My Second Offline Eatblock Sync Test is complete (with DB Read Wait Stats): http://www.mod6.net/eatblock-test/
mod6: I have updated all the statistics, charts, and commentary in place.
mod6: Thanks to diana_coman for the help with gnuplot!
mircea_popescu: o hey
mircea_popescu: mod6 maybe i'm thick, but where's the per-line timing per block ?
mod6: hi
mod6: if you're looking at this chart: http://www.mod6.net/eatblock-test/trb_offline_eatblock.png
mod6: this is built from this line in the debug.log, for example: "ProcessBlock (res == 1) took : 2901ms; db write wait: 313ms; db read wait: 155ms "
mod6: the red lines are the first number after "took", so in the case of the example above, ProcessBlock took 2901 milliseconds.
mod6: is that what you're asking?
mod6: It wouldn't take much more effort to add the AcceptBlock values into the trb_offline_eatblock.png if that's wanted.
mircea_popescu: what i was looking for was, supposing the whole of trb looks like : http://wotpaste.cascadianhacker.com/pastes/YOK7i/?raw=true then something like : http://wotpaste.cascadianhacker.com/pastes/qGO07/?raw=true
mircea_popescu: literally, time per line of code per block.
mod6 looks
mod6: ah. ok.
mircea_popescu: kinda give a time heatmap for the whole codebase, as it were.
mircea_popescu: asciilifeform "specifically made to" means a lot less than it appears to, for reasons such as euler's equality etc. pi is a fundamental constant, it's everywhere.
mod6: sure. i missed that point completely - was just putting some stats together from alf's timer/odometer vpatches. i think something along the lines of what you're referring to takes deeper, or much more extensive instrumentation / profiling.
mod6: asciilifeform: unless I'm missing something major from one of your vpatches? (i've also posted my debug.log -- please take a second to see if I'm missing something major if you have a spare one)
mod6: but, in any case, I could work on further analysis like that, mircea_popescu

2017-4-25