[tor-talk] (no subject)

Nick Mathewson nickm at alum.mit.edu
Mon Aug 17 19:27:01 UTC 2015


On Tue, Aug 11, 2015 at 9:28 PM, Thomas White <thomaswhite at riseup.net> wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Does anyone in Tor want to name a price to get this task done? Can
> then be followed by a match donation to be spent with on whatever you
> wish once the multicore has been added.

Hi, Thomas, and congratulations!  You've asked a question I wasn't
prepared to answer.  Here's a thread we had about it today:

18:25 < nickm> So, I assume people have seen the tor-relays/tor-talk
thread about "Hey Tor folks, what would you want in exchange for
making tor parallelize better"
18:25 < nickm> Do we have a way of even answering that?
18:26 < nickm> If not, I think we should reply to say "This is a
generous offer and we need to apologize for taking so long, but it's
not been something we had a way of answering before. We'll try to come
up with such a way and see what it outputs RSN"
18:26 < nickm> thoughts?
18:44 < arma4> sounds plausible. i think the issue is a combination of
not enough developerpower and also not enough money
18:44 < arma4> a short small amount of money wouldn't be enough to
overcome the first issue,
18:45 < arma4> and we need to overcome both
18:47 < nickm> yeah.  I think that anything less than a year fulltime
of dev time, plus overhead and incidentals, can't work out here.
18:47 < nickm> plus, no timeline promised
18:48 < nickm> arma4: thoughts?
18:49 < arma4> are there any incremental steps that can be done, by
other people, in the mean time?
18:49 < nickm> in theory sure
18:49 < arma4> it seems like a wildly unpredictable amount of work
18:50 < nickm> in practice nobody who isn't a Solid Wizard is going to
get much done here
18:50 < arma4> and it's not even clear, to me, what architecture we
should use to parallelize cleanly
18:50 < arma4> all of this ipc stuff sounds great in theory until you
try to run the program on ios or something and then boy are you
surprised
18:51 < nickm> I have an architecture in mind for circuit crypto
18:51 < nickm> for tls, I have no bloody clue
18:52 < toml> but would we feel good about taking a shot if there was
one full-time equivalent devoted to the problem?
18:52 < arma4> maybe explaining very briefly why it isn't trivial, and
why it is going to be hard to do right, would be helpful for the folks
wondering why we don't just do it already
18:52 < toml> arma: I agree that we should take the opportunity to
explain the challenge
18:52 < arma4> toml: and if we had said full-time developer, would
this be the most important thing to have her work on?
18:53 < arma4> so far the answer has been "no, other things are more important"
18:53 < toml> well, it would be an answer to the question: what would it take?
18:53 < toml> so if they put cash on the barrel head, we could
dedicate. (I would bet there would be other associated benefits not
strictly related)
18:54 < toml> probably the cost would be too steep, but they would
know where we stand. (part of the education piece)
18:55  * nickm suggests that we just copy-and-paste this conversation
into the thread
18:56 < arma4> sounds good
18:58 < toml> arma: and let's always use the term "full-time
equivalent." There is an industry standard for a FTE amount, but we
reserve the right to apportion those funds among more than one person.
18:58 < nickm> any more to add ?
19:00 < nickm> I feel like we could safely say "More than 80k and less
than 500k" on this today, and if those numbers don't scare people
away, invest time into digging into getter numbers
19:01 < arma4> sounds good. it is basically a big architectural change
inside tor. our work on better testing and better modularity is
(slowly) moving us in the right direction as we wait.
19:01 < toml> I would say minimum $100K, as this would leapfrog
several other priorities.
19:02 < nickm> also overhead
19:02 < toml> si
19:02 < nickm> yeah, good point, toml
19:03 < nickm> OTOH, we can also mention the $0 price point: for no
money at all, we will _care_ about this, because we already do. And at
some point eventually, somebody will surely work on it in their free
time, one of these days
19:03 < toml> (and that is too low for a FT equivalent, but it is
enought to motivate us to explore
19:03 < arma4> heck, not only do we care, but we even wrote up a thing
on how it might be done
19:04 < toml> arma: should we share that? (or share it again?)
19:05 < arma4> nickm should point to it in his response i hope
19:06 < arma4> he wrote it so hopefully he knows what is the best
thing to point at :)
19:06 < nickm> well,it's quite old and maybe I should revise some
morning/afternoon when I am smarter
19:08 < toml> perhaps leave it as is — show how long we have been
thinking on this. Then maybe a add brief bit on things we have learned
since, at your leisure.
19:08 < arma4> that way lies paralysis. which is almost like
parallelization, but not quite. :)
18:13 < nickm> ok.  So I am going to add this to topics for the
wednesday core tor dev meeting, and send it to the ml, unless somebody
objects?

I think the URL I was asked to add was
https://trac.torproject.org/projects/tor/wiki/org/projects/Tor/MultithreadedCrypto
, but that's rather outdated.

tl;dr: more data really soon now. We are bad at doing cost estimates
of this kind, and hope to get better RSN. Thanks for the interest!


peace,
-- 
Nick


More information about the tor-talk mailing list