[hashcash] Re: Sample code for Dworkian memory-bound POW

  • From: Jonathan Morton <chromi@xxxxxxxxxxxxxxxxxxxxx>
  • To: hashcash@xxxxxxxxxxxxx
  • Date: Wed, 26 May 2004 00:14:19 +0000 (UTC)

> You need to create a file mempow.dat of 16 MiB + 1 KiB in size.
> As described in the comments I concatenated two 10 MB files from
> random.org, but for timing purposes you can probably use anything.
> Just copy a file that is > 16 MB in size to mempow.dat.

I assume that if implemented, every sender needs to have the same file.  That
would result in very big download sizes for mail clients.  That's a huge barrier
to adoption.
 
There's a further general problem with memory-limited algorithms:  it's
impossible to make an algorithm which is small enough for consumer machines
while still being too big to be optimised by a high-end workstation.

There are still a huge number of Win98 machines out there which have barely
enough RAM to load the OS, and would be crippled by an additional 16MB block
which was needed to send e-mail.

I also have a real example of low-end Macs - often with only 8MB total RAM -
being distributed to deprived individuals to keep them in touch with relatives
and support organisations.  Such old machines would be totally incapable of
running a Dworkian algorithm, rather than merely being slow as with a hash-based
algorithm.

At the same time, it's technically plausible to build a high-end workstation
with 16MB of high-speed, low-latency cache, which would be at a massive
performance advantage compared to a DRAM-limited system.  Such machines may
already be available - the IBM POWER4 processor module contains a total of 8MB
of shared L2 cache, and I haven't reviewed IBM's POWER5 workstation specs yet.

Even if not, a determined spam ring could easily build such a machine
themselves, based on, say, an ARM MPCore processor bank and an SRAM-based memory
array.  Once designed, that machine would not be greatly more expensive to
produce (in orders of magnitude, that is) than a headless PC, but could be as
much as 50x as fast.



Other related posts: