On Fri, 6 Aug 1999, athompso wrote:
>Well, from a theoretical standpoint, an Alpha flavour of chip running
>at, say 500Mhz should have comparable floating-point performance to an
>Intel flavour of chip running at the same internal clock rate. The
>Alpha chip should have an edge of no more than 50% on a per-CPU basis.
>Architechtural issues should only account for 5% or so (ie. inter-CPU
>bus & cache design).
There can be larger differences than this, depending on what you are
doing, and exactly which model of Intel and Alpha you have. A difference
of more than a factor of 4 at the same clock speed would be surprising
however. Normally Intels are at their worst and Alphas at their best on
floating point, and programs that use large memory areas intensively.
>SSL is, as far as I understand, highly floating-point intensive, as it
>must (typically) calculate RSA algorithms in software. The Alpha chip
>should excel in this area. But not by an order of magnitude.
>-Adam
RSA algorithms involve large integer calculations, not floating point. I
haven't looked at the large integer code in SSL. I have looked at
old code by Lenstra and Manasse. It used no floating point in the
inside loops. It did a floating point divide in the next loop out of some
routines, but i don't think those were the ones needed by the RSA
algorithm. I have also helped implement a large integer arithmetic library
that used no floating point operations at all and performed well. I can't
figure out how to make heavy use of floating point in the RSA algorithm.
______________________________________________________________________
Apache Interface to OpenSSL (mod_ssl) www.modssl.org
User Support Mailing List [EMAIL PROTECTED]
Automated List Manager [EMAIL PROTECTED]
