[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Date: Tue, 9 Sep 2014 15:50:03 +0400
From: Solar Designer <solar@...nwall.com>
To: discussions@...sword-hashing.net
Subject: BSTY - yescrypt-based cryptocoin
Hi,
A yescrypt-based cryptocoin was launched ~12 hours ago:
https://bitcointalk.org/index.php?topic=775289.0
https://github.com/GlobalBoost/GlobalBoost-Y
https://bst.globalboo.st/y/product.php?id=22
They currently credit me for "recommendations", but actually my
involvement was minimal, and I'd like to keep it minimal, so I am asking
them to remove that wording (hopefully, it'll be gone by the time you
read this). I am not entirely happy with this cryptocoin, but I think
it's a fine way to test yescrypt better if the coin ends up providing
incentive for breaking/optimizing yescrypt.
Currently, the only miner is the wallet. As released, it uses
yescrypt's SIMD-less code (with a minor incompatible change possibly
inadvertently introduced by BSTY developers) and it has memory
(de)allocation overhead in the loop.
The attached patch introduces the change/bug into yescrypt's SIMD code
in the wallet tree and uses thread-local storage (yes, a hack, and
not universally supported) to easily move the memory (de)allocation
overhead out of the loop. Someone may reimplement the latter in a
better way (without a dependency on thread-local storage). Of course,
the official yescrypt won't include these changes; they are BSTY-only.
The official yescrypt's API allows for the caller to keep the memory
(de)allocation overhead out of the loop, but this wallet code didn't use
that. And it was quicker for me to hack than to implement such use.
Before this patch, BSTY mining ran at 1300 hashes/s on i7-4770K. With
the patch, it's the expected 3400 hashes/s (same as my "userom 0 2"
benchmark reports for 8 threads). However, the speed drops to zero
whenever the wallet loses connection to network, which happens very
often (perhaps overloaded nodes).
BSTY uses yescrypt with: N=2048, r=8, p=1, t=0, (YESCRYPT_RW |
YESCRYPT_PWXFORM). The parameters are fixed (no growing N, unlike e.g.
in YACoin).
Luckily, BSTY's incompatible change to yescrypt does not affect its
security, nor its performance. So BSTY can serve to test yescrypt's
important properties despite of this change.
I am BCC'ing this to a BSTY developer.
Alexander
View attachment "GlobalBoost-Y-fast.diff" of type "text/plain" (3805 bytes)
Powered by blists - more mailing lists