Re: Performance Comparison

[Phillip Ezolt <ezolt@perf.zko.dec.com>]

Well,
	If any one has an alpha system availble, Iprobe has been ported to 
FreeBSD and Alpha/Linux.  They would be able to do a more indepth analysis than
the following:

+---------------------------------+
| Profile Report | Event = Cycles |
+---------------------------------+

    * One sample = 65536 events *


Begin            End                                    Sample Image Total
Address          Address          Name                   Count   Pct   Pct
-------          -------          ----                   -----   ---   ---
0000000000000000-0000000120006F2F /root/hint/DOUBLE    4363066        95.7
0000000120001800-0000000120001B1F   Hint               4355514  99.8  95.6
FFFFFC0000300000-00000000FFFFFFFF vmlinux               186636         4.1
FFFFFC0000315E40-FFFFFC0000315F7F   do_entInt            79440  42.6   1.7 


+--------------------------------------+
| Profile Report | Event = Bcache Miss |
+--------------------------------------+

    * One sample = 16384 events *


Begin            End                                    Sample Image Total
Address          Address          Name                   Count   Pct   Pct
-------          -------          ----                   -----   ---   ---
0000000000000000-0000000120006F2F /root/hint/DOUBLE      34579        93.6
0000000120001800-0000000120001B1F   Hint                 34571 100.0  93.6
FFFFFC0000300000-00000000FFFFFFFF vmlinux                 2232         6.0
FFFFFC0000315E40-FFFFFC0000315F7F   do_entInt              768  34.4   2.1


(I asked Iprobe to only show me functions where >1% of the time was being
spent)

We can see that time is not being spent in any libraries.

We can get a breakdown by PC, but that isn't particularly useful unless
someone really wants to take time to look at it. 

Iprobe should only add a small amount of noise to the system.  (5%)

--Phil

Digital/Compaq:                     HPSD/Benchmark Performance Engineering
Phillip.Ezolt@compaq.com                            ezolt@perf.zko.dec.com


On Tue, 4 May 1999, Alan Cox wrote:

> > Alan probably misread the original message because HINT benchmark indeed
> 
> I did. Sorry for not replying earlier but I've been in the Algarve for a few
> days.
> 
> > As with any other benchmark the compiler used does show in the results
> > but not terribly much. To be completely certain the benchmark should be
> > compiled with the same compiler, version and optimization flags which I
> > think wasn't the case with Otto Solares' test runs but this can't
> > explain all the differences.
> 
> The BSD and Linux gcc will produce close to identical code and different
> libcs. If libc is involved then its still clearly a valid benchmark just
> different things need fixing
> 
> > Yes, I agree. I suggest more people to try the HINT benchmark. The code
> > is ANSI-C and seems generally professional quite unlike the bytemark
> > benchmark.
> 
> A benchmark from an outside source, with no bias, with source code and no
> huge fees is a rather nice thing. 
> 
> Richard Gooch has a performance monitoring patch for Linux. The PMC registers
> on the PPro and higher can count many types of cache misses which may be
> most enligtening
> 
> 
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.rutgers.edu
> Please read the FAQ at http://www.tux.org/lkml/
> 
> 


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/