glibc with NPLT support
Dennis Veatch
dveatch at woh.rr.com
Wed Dec 7 20:57:01 UTC 2005
On Wednesday 07 December 2005 11:02, Florin Braescu wrote:
> Hi everyone,
>
> In the last few mounths i have installed and used on my system an NPLT
> enabled glibc module, instead of the one provided by the Lunar Linux. From
> my experience with it i can say that the system seems to be faster and
> smoother. Im attach here the module i use for those people which want to
> test this option. I do not recommend it to be installed on production and
> critical systems. It is not widely tested to guarantee that no problems can
> occur. The installation is simple, but time consuming. You must install
> this module first, relin gcc, then relin this module again (with the new
> compiled gcc). After that a lunar rebuild can be done in order to create
> modules which use this option and to achieve the maximum performance. I am
> very interested of your experience with this module. Have fun.
>
> Yours,
>
> Florin
Tried it out and did a lunar rebuild. No gremlins or gotchas reared up... so
that's good. I didn't think about it till well into the rebuild that some
kind of benchmark should had been done. Oh well. So anything I have to say is
purely subjective and that would be on the good side.
Having said that. I did run nbench with the only remaining things needing a
rebuild was KDE. So while it was working on kdebindings3 during the make
process (it always seems to grind along for me) ran nbench from within KDE.
That's the first set of numbers below. The second set is (from within KDE)
and no lin process going on.
I've never run nbench before and not at all familiar with it but I thought the
results interesting anyway. There was afaiac no difference in performance. So
either nbench suspends other tasks while it's running or there just might be
something to this nptl thingy.
One thing nbench doesn't show is the amount of system ram which is 2G, DDR400.
Oh and uh, 32 bit mode.
TEST : Iterations/sec. : Old Index : New Index
: : Pentium 90* : AMD K6/233*
--------------------:------------------:-------------:------------
NUMERIC SORT : 1281 : 32.85 : 10.79
STRING SORT : 140.5 : 62.78 : 9.72
BITFIELD : 4.2846e+08 : 73.50 : 15.35
FP EMULATION : 102.92 : 49.39 : 11.40
FOURIER : 22598 : 25.70 : 14.44
ASSIGNMENT : 18.464 : 70.26 : 18.22
IDEA : 3886 : 59.44 : 17.65
HUFFMAN : 1383.9 : 38.38 : 12.25
NEURAL NET : 29.848 : 47.95 : 20.17
LU DECOMPOSITION : 1146.6 : 59.40 : 42.89
==========================ORIGINAL BYTEMARK RESULTS==========================
INTEGER INDEX : 53.159
FLOATING-POINT INDEX: 41.831
Baseline (MSDOS*) : Pentium* 90, 256 KB L2-cache, Watcom* compiler 10.0
==============================LINUX DATA BELOW===============================
CPU : AuthenticAMD AMD Athlon(tm) 64 Processor 3500+ 2211MHz
L2 Cache : 512 KB
OS : Linux 2.6.15-rc2
C compiler : gcc version 3.4.4
libc : libc-2.3.6.so
MEMORY INDEX : 13.957
INTEGER INDEX : 12.770
FLOATING-POINT INDEX: 23.201
Baseline (LINUX) : AMD K6/233*, 512 KB L2-cache, gcc 2.7.2.3, libc-5.4.38
TEST : Iterations/sec. : Old Index : New Index
: : Pentium 90* : AMD K6/233*
--------------------:------------------:-------------:------------
NUMERIC SORT : 1280 : 32.83 : 10.78
STRING SORT : 140.66 : 62.85 : 9.73
BITFIELD : 4.2918e+08 : 73.62 : 15.38
FP EMULATION : 102.88 : 49.37 : 11.39
FOURIER : 22177 : 25.22 : 14.17
ASSIGNMENT : 18.449 : 70.20 : 18.21
IDEA : 3890.7 : 59.51 : 17.67
HUFFMAN : 1386.7 : 38.45 : 12.28
NEURAL NET : 29.797 : 47.87 : 20.13
LU DECOMPOSITION : 1140.8 : 59.10 : 42.68
==========================ORIGINAL BYTEMARK RESULTS==========================
INTEGER INDEX : 53.190
FLOATING-POINT INDEX: 41.474
Baseline (MSDOS*) : Pentium* 90, 256 KB L2-cache, Watcom* compiler 10.0
==============================LINUX DATA BELOW===============================
CPU : AuthenticAMD AMD Athlon(tm) 64 Processor 3500+ 2211MHz
L2 Cache : 512 KB
OS : Linux 2.6.15-rc2
C compiler : gcc version 3.4.4
libc : libc-2.3.6.so
MEMORY INDEX : 13.966
INTEGER INDEX : 12.776
FLOATING-POINT INDEX: 23.003
Baseline (LINUX) : AMD K6/233*, 512 KB L2-cache, gcc 2.7.2.3, libc-5.4.38
* Trademarks are property of their respective holder.
--
You can tuna piano but you can't tune a fish.
More information about the Lunar
mailing list