phc-discussions - Re: [PHC] Re: Initial hashing function. Feedback welcome

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 linux-cve-announce 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: Thu, 2 Jan 2014 11:30:58 -0500
From: Bill Cox <waywardgeek@...il.com>
To: discussions@...sword-hashing.net
Subject: Re: [PHC] Re: Initial hashing function. Feedback welcome

I think how much memory a memory-hard KDF can fill per second is likely
it's most important performance metric.  However, to really benefit from
memory-hard KDF, you need to run for longer than many servers can afford.
 This is why I think client side KDF is important.

A maxed out server may only have 1ms per user for KDF to spare.  There's
some advantage in maxing out memory vs doing SHA-256 even at 1ms, but it's
not large.  As you hash longer with a memory hard function, the cost to an
attacker goes up because he'll need to match your memory in addition to
running longer.  Just doing more SHA-256 hashes makes him run longer, but
does not increase his hardware cost.  By the time you hash for a second,
the memory hard KDF has a huge cost penalty to an attacker who builds
custom ASICs.  That cost penalty is proportional to the bandwidth of the
KDF, which is why this metric is so important.

We can increase memory bandwidth with multiple threads, up to a point.  On
my machine, that point seems to be 2 threads for my current version of
keystretch.  After that, there's little point in having more CPUs involved,
IMO.  Maybe we could have them do SHA-256 hashes, but that wont
significantly increase the cost to an attacker.

On Thu, Jan 2, 2014 at 10:02 AM, Peter Maxwell <peter@...icient.co.uk>wrote:

>
>
>
> On 2 January 2014 14:44, Bill Cox <waywardgeek@...il.com> wrote:
>
>>
>>
>> Running multiple copies of memmove shows my system seems to handle about
>> 23GB/s of memory bandwidth.  Two copies of keystretch achieve about 13GB/s.
>>  It's not maxing out memory bandwidth, but it's getting there...
>>
>
> Surely you do not want to max-out memory bandwidth, that's more a problem
> than a benefit is it not?  For example, password checks are but a small
> proportion of what an average web-server must do.
>
>
>
>

Content of type "text/html" skipped