Date: Thu, 17 Apr 2014 13:42:01 -0300 (BRT)
From: Marcos Antonio Simplicio Junior <mjunior@...c.usp.br>
To: discussions@...sword-hashing.net
Subject: Re: [PHC] Lyra2 initial review

Bill and Alexander, 

----- Mensagem original -----

> De: "Solar Designer" <solar@...nwall.com>
> Para: discussions@...sword-hashing.net
> Enviadas: Quinta-feira, 17 de Abril de 2014 13:21:33
> Assunto: Re: [PHC] Lyra2 initial review

> Bill,

> How did you calculate the "average memory/second" for Lyra2 and
> yescrypt. I don't know about Lyra2's, but yescrypt's looks wrong to
> me.

Actually an explanation would be really good, because I confess that, except for "memory bandwidth", I'm not sure I understand what those numbers mean or how to computed them... 

> I think it is 5/8 of peak, since memory usage is growing linearly for
> the first 3/4th of time, and then stays at peak for 1/4th.

> On Thu, Apr 17, 2014 at 07:56:55PM +0400, Solar Designer wrote:
> > On Thu, Apr 17, 2014 at 11:08:25AM -0400, Bill Cox wrote:
> > > Lyra2:
> > > Peak memory/second: 1.63 GiB/s
> > > Average memory/second: 1.22 GiB/s
> > > Memory bandwidth: 11.4 GiB/s (highly bandwidth limited)
> > >
> > > Yescript:
> > > Peak memory/second: 3.03 GiB/s
> > > Average memory/second: 2.27 GiB/s

> This would be: 3.03*5/8 = 1.89 GiB/s

> > > Memory bandwidth: 6.06 GiB/s (does it do 2 r/w to DRAM per
> > > location, or 3?)
> >
> > yescrypt does 4 r/w per location when YESCRYPT_RW is set (and it is
> > set

> No, that's wrong: it's 2 r/w per iteration. It would be 4 r/w per
> location if SMix2 ran for as many iterations for SMix1, but in this
> mode
> it does not. I think correct is:

> 3.03 * 4/3 * 2 = 8.08 GiB/s

> 4/3 is ratio of number of iterations to number of locations
> 2 is number of r/w per iteration

> > in the PHS() API), so this would be 12.12 GiB/s. It does one random
> > read and one sequential write per iteration in SMix1, and one
> > random
> > read+write per iteration in SMix2. (Resetting, YESCRYPT_RW brings
> > it to
> > only 2 r/w per location like in normal scrypt, but then it'd have
> > to run
> > 1.5 times longer to achieve optimal normalized area-time and it'd
> > be
> > friendly to TMTO, hence that mode isn't the default.)
> >
> > > TwoCats:
> > > Peak memory/second: 4.98 GiB/s
> > > Average memory/second: 2.49 GiB/s
> > > Memory bandwidth: 9.97 GiB/s
> >
> > I'd interpret these as TwoCats and yescrypt winning over Lyra2 in
> > terms
> > of average memory/second, which I think is the most important of
> > three
> > metrics here. yescrypt loses to TwoCats in terms of this metric by
> > 8.9%,
> > but it possibly makes up for that by using 21.6% more memory
> > bandwidth
> > (some would view that as a drawback, but in this test we were
> > specifically
> > trying to use as much bandwidth from one thread as we could).

> The 8.9% was based on your "average memory/second" figures. With
> yescrypt's corrected, TwoCats' advantage is 24%, which is impressive!

> I am surprised that Lyra2 and TwoCats seem to win significantly in
> terms
> of bandwidth usage. I doubt it's the cost of one round of pwxform
> (although to find out we can try making it zero rounds, even). It
> could
> be the cost of r=8.

It is likely to have a relationship: we have a compile-time parameter "nCols" that has a similar effect to scrypt's "r". It is part of the algorithm's core rather than a "tweak", but it would not fit in the PHS interface... 

> > BTW, chances are that going from r=8 to r=16 (already a tunable,
> > like in
> > scrypt) would increase yescrypt's bandwidth usage (and average and
> > peak
> > memory usage per second) some further. I choose r=8 for PHS()
> > because
> > it increases reliance on fast small random lookups. r=32 and higher
> > might be faster or slower, since we're competing for L1 cache space
> > with
> > pwxform's S-boxes (8 KiB in these tests).

> Alexander

Content of type "text/html" skipped

Powered by blists - more mailing lists