[cap-talk] FW: x.509 -- MD5 considered harmful today

[John Carlson]

On Dec 31, 2008, at 3:44 PM, zooko wrote:

> On Dec 31, 2008, at 15:20 PM, Bill Frantz wrote:
>
>> In my experience, the code which performs the I/O is frequently
>> expensive enough to be a bottle neck. Many compression functions
>> can pay for themselves by reducing the CPU cost of the I/O. How do
>> these hash functions compare in CPU cost with common compression
>> functions?
>
>
> Secure hash functions are more expensive.  SHA-256 is about 21 cycles
> per byte.  The fastest open source compression (multi-algorithm, type-
> detecting) that I know of -- FreeARC -- is about 3.7 cycles per
> second (on a standard test set).  zlib is about 16 cycles per byte.
>
> I'm hoping that SHA-3 will turn out to be substantially faster than
> SHA-256, but it might not be.

Are any of these algorithms parallelizable?  Say I had 256 processors.

John