GPG & Floating point?

[Bernd Eckenfels]

On Thu, Dec 20, 2007 at 01:33:07PM +0100, Werner Koch wrote:
> Alan wrote:
> > I'm running some performance tests on a 4-core SPARC T-2000 verses my 
> > current production environment, and GPG decryption performance is 
> > nowhere near as improved as I would expect.  The actual performance I 
> > get per thread is about equal to an old 296 MHz Ultra-SPARC II 

On T1/T2 it is possible to see the number of FP operations with cpustat:
http://developers.sun.com/solaris/articles/t1t2_perf_counter.html#FloatingPointComputation

Besides that, a T1 single core is pretty slow, and as long as the code is
not optimized for using the hardware accelerator it might really look very
slow.

> It depends on what you are measuring: Is it about bulk operation
> (processing large files) or the performace of the public key operations
> on short data.  The latter can be improved by using the assembler code
> for the big integer math; check out what links are set in mpi/.  Bulk
> encryption is limited by the encryption code and the SHA-1 hash
> function.  Are you using the same algorithms on both machines?

Exactly. I wonder if there is platform code to use the T2000 crypto hardware
accelerator for DSA/RSA: http://www.sun.com/servers/coolthreads/tnb/faq.jsp#q_10

(not sure if there is any way to speed up hashing besides hand crafting
assembler for the T1).

Gruss
bernd
-- 
  (OO)     -- Bernd_Eckenfels at Mörscher_Strasse_8.76185Karlsruhe.de --
 ( .. )    ecki@{inka.de,linux.de,debian.org}  http://www.eckes.org/
  o--o   1024D/E383CD7E  eckes at IRCNet  v:+497211603874  f:+49721151516129
(O____O)  When cryptography is outlawed, bayl bhgynjf jvyy unir cevinpl!