lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  PHC 
Open Source and information security mailing list archives
 
Hash Suite for Android: free password hash cracker in your pocket
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Wed, 10 Sep 2014 07:15:43 +0400
From: Solar Designer <solar@...nwall.com>
To: discussions@...sword-hashing.net
Subject: Re: [PHC] BSTY - yescrypt-based cryptocoin

On Tue, Sep 09, 2014 at 03:45:28PM -0400, Bill Cox wrote:
> On 09/09/2014 01:36 PM, Bill Cox wrote:
> > I just started it up on my testing server (yes, my son's MineCraft 
> > server).  With your patch it does up to 3,100-ish hashes/second, 
> > slightly below what your processor does.  It has mined 1,388 BSTY
> > so far.  I'm rich! :-P

Sorry to disappoint you, but it appears they have another nasty bug in
there (and probably more).  It looks like some systems (all 64-bit?)
mine a different blockchain from 32-bit ones (which are apparently the
majority, due to Windows binary wallet).  I reported it and provided
some detail, so I hope they'll figure this out and deal with it somehow.

I did verify, after they already launched the coin, that on my 64-bit
Linux build their wallet's integrated yescrypt code produces the same
hashes as my standalone yescrypt code does after I introduced their
now-known bug to the SHA-256 call (uses salt instead of "Client Key").
So it appears the new bug, yet to be discovered, is somewhere outside of
the yescrypt code in the wallet, yet somewhere within the wallet code
tree.

I notice they cast pointers to structs containing the wallet's uint256
class to (char *) when invoking yescrypt, and I'm told "others do the
same", yet I suspect this may be the culprit.  It'd surely fail on
little- vs. big-endian, as it would be too inefficient to implement
uint256 as having same layout despite of underlying integer types'
endianness.  I guess it might fail for 32-bit vs. 64-bit "long" as well.

Also, I doubt all of the many people having edited the wallet code are
aware of C strict aliasing uses, but luckily (char *) is a special case.
With the different-typed uses from different source files and without
link-time optimization, this should happen to work as far as C strict
aliasing violations are concerned.

Whatever.  Dirty stuff, and not enough pre-launch testing.

> I did a quick comparison of TwoCats and Yescrypt when doing 2MiB
> hashes.  Yescrypt maxes out my machine at about 3,100 hashes per
> second using 8 threads, which gives the best performance.  TwoCats
> maxes out at about 3,800 similar sized hashes on 3 threads with 2
> multiplications per inner loop, which gives the best performance.
> However, Yescrypt is doing something like 2.3 memory read/writes per
> location vs TwoCat's 2.  The difference is basically in the noise.

Yeah.  To add one more number, yescrypt with pwxform rounds count
reduced from 6 to 2 does 4400 hashes per second on the same i7-4770K
where it does 3400 hashes per second with the default of 6.  I guess
this might correspond to more than 3800 on your machine. ;-)  But I
think the default usually works better, since it provides much more
compute latency hardening:

6*3400/(2*4400) = 2.3 times more hardening

BTW, did you enable AVX when building yescrypt?  This wasn't part of the
BSTY wallet patch I posted (need to add -mavx or -march=native).  Not
that it makes a lot of a difference on Ivy Bridge, though, where SIMD
MOVs are implemented via register renaming (IIRC), but it might help a
bit.  Optimal code for yescrypt does not need any MOVs within pwxform
round on plain SSE* anyway.  It's just that compilers may produce some.

> I wonder about the choice not to bust into external DRAM.  This size
> hash could fit between 4 and 8 2MiB cache RAMs on a high-end ASIC
> (same process as my CPU).  Had they used 32MiB or more, the ASIC might
> need high-speed external DRAM interfaces, and these things are tricky
> to get right, significantly complicating an ASIC effort.

Yeah, but unfortunately verification time is a concern for cryptocoins.

Yes, YACoin is already at 8 MiB and growing, but I would not be
surprised if eventually the verification times for it become
unacceptable and it has to be abandoned or forked, especially as the
number of transactions grows.  I am no expert in this, though.  With no
expert involved, 2 MiB for BSTY felt like a safer bet.  Besides, I am
interested in having yescrypt at 2 MiB attacked, as that's the setting
relevant for some other uses. ;-)

This leaves room for improvement in another yescrypt-based coin, e.g.
introducing very slowly growing N (much slower than in YACoin,
especially considering that YACoin's trivial modification of scrypt is
TMTO-friendly and yescrypt in native mode is not).

> Still, there's probably no more than about a 10X speed improvement for
> a very high-end ASIC vs my CPU (4 cores running 25% faster).  That's
> really fantastic ASIC defense.  I wouldn't expect to see such a
> high-end ASIC for this application for a long time.

We'll see.  I agree there may eventually be ASICs for yescrypt at this
memory size.  With a fast enough bus, they may even use external DRAM to
pack more 2 MiB instances per ASIC than would otherwise fit on die.

> It is possible
> that this system will succeed at distributing the hashing load (and
> mining profits) evenly among users, which would be very cool.  It
> seems a lot more fair to let anyone running a client be on essentially
> the same level as everyone else.

The usual concern here is botnets.  But maybe that's sort of fine:
provides incentive to use botnet nodes' computing power for mining
rather than e.g. for password cracking. ;-)

Alexander

Powered by blists - more mailing lists