[rc5] Difference between SunOs-binaries...

Dirk Kuypers dk at comnets.rwth-aachen.de
Thu Nov 6 14:06:37 EST 1997


On Thu, 6 Nov 1997, Jeff Lawson wrote:

> At 12:55 PM 11/6/97 +0100, Dirk Kuypers wrote:
> >what's the difference between the two SunOs-clients besides the release 
> >dates:
> >
> >SunOS 4.1.x (Sparc rolled) (new 11/6) 
> >SunOS 4.1.x (Sparc) (new 10/30) 
> 
> 
> Just an experiment of rolled vs. unrolled code.  The rolled version might
> be faster than the unrolled (normal) one on old machines with especially
> small instruction caches.  Just benchmark the two versions and see which is
> fastest for you.

Ah, ok,

then here come some benchmarks, maybe they are usefull to others:


Sparc Classic, 50 MHz:

*** unrolled code: ***

Benchmarking with 10000000 tests:
.....10%.....20%.....30%.....40%.....50%.....60%.....70%.....80%.....90%....
Complete in 644.67 seconds. [15511.61 keys/sec]

*** rolled code ***

Benchmarking with 10000000 tests:
.....10%.....20%.....30%.....40%.....50%.....60%.....70%.....80%.....90%....
Complete in 654.76 seconds. [15272.61 keys/sec]

-----------------------------------------------------------------------
SS5, 110 MHz:

*** unrolled Code: ***

Benchmarking with 10000000 tests:
.....10%.....20%.....30%.....40%.....50%.....60%.....70%.....80%.....90%....
Complete in 114.11 seconds. [87628.41 keys/sec]

*** rolled code: ***

Benchmarking with 10000000 tests:
.....10%.....20%.....30%.....40%.....50%.....60%.....70%.....80%.....90%....
Complete in 114.88 seconds. [87040.49 keys/sec]

----------------------------------------------------------------------------

SS10, 20 MHz:

*** unrolled code: ***

Benchmarking with 10000000 tests:
.....10%.....20%.....30%.....40%.....50%.....60%.....70%.....80%.....90%....
Complete in 180.52 seconds. [55393.25 keys/sec]

*** rolled code: ***

Benchmarking with 10000000 tests:
.....10%.....20%.....30%.....40%.....50%.....60%.....70%.....80%.....90%....
Complete in 183.52 seconds. [54487.38 keys/sec]

HMM, at least here I can see no big difference between the two clients. I 
think the differences , 1kkeys/sec are more related to different load 
during the benchmarks.

Dirk

--
  Dirk Kuypers                  ComNets - RWTH Aachen
                                 
       

----
To unsubscribe, send email to majordomo at llamas.net with 'unsubscribe rc5' in the body.



More information about the rc5 mailing list