Re: additional core draft nits in need of WG attention.

[nisse%lysator.liu.se@localhost (Niels Möller)]

Bill Sommerfeld <sommerfeld%east.sun.com@localhost> writes:

> - what proportion of the code size of a "minimized" implementation is
> the actual algorithm code?  i.e., are we talking 1% or 10% bloat here?

In my implementation, des is about 8K code. Aes is about 2.5K code and
9K tables, which iirc can easily be cut to about 3K of tables if
memory usage is more important than speed, so below I'll count aes as
5.5K total code and data. All this measured on x86.

The stripped lsh client binary is 340K (excluding bignum library and
libz and other libraries which are linked dynamically). So here, the
ciphers are 2.4% (des), or 1.6% (small aes) of the bulk.

So let's guess that a minimized implementation would be cut down to
somewhere between 100K-200K (I think it's difficult to get below 100K,
anybody that has tried?). Then the bloat factor for each cipher is
somewhere between 3% and 8%.

/Niels