From:	DJ Delorie <dj AT delorie DOT com>
Subject:	Re: B19, performance using fopen()/fclose() on FAT16
24 Dec 1998 13:08:32 -0500 :
Message-ID:	<36828320.893CEDC2@delorie.com>
References:	<3680FDDC DOT 5923BA32 DOT cygnus DOT gnu-win32 AT aud DOT alcatel DOT com>
Mime-Version:	1.0
X-Mailer:	Mozilla 4.5 [en] (X11; I; Linux 2.1.126 i586)
X-Accept-Language:	en

John Blanton wrote:
> 
> Conducting performance tests of several porting tools we noticed
> dramatically poor performance for the following case using FAT16 on
> Windows NT:
>
> REPETITIONS is set to 100,000.  The test requires over an hour to
> complete.  There is constant disk activity.  Here are some test timings:
> 
> GNU-Win32           398.3
> Linux (FAT)           6.2
> Linux (Linux FS)      6.1
> U/WIN                41.0
> Win32                14.8

I performed the same tests with B20.1 and djgpp on a dual PII/300.
Sorry, don't have Linux on those systems.  Numbers are microseconds
per loop.  Sorry, can't compare these to your numbers without
matching hardware, but please compare your results to the same
program built with cygwin B20.1 (or djgpp 2.02) and see if you
notice a difference.

win95	djgpp	fat16	1380 us
win95	cygwin	fat16	1500 us
winnt	djgpp	fat16	 360 us
winnt	cygwin	fat16	 890 us (89 seconds for 100,000 calls)
winnt	djgpp	ntfs	 400 us
winnt	cygwin	ntfs	 530 us

Note: I *expect* cygwin to be slower than djgpp, because it
has to simulate the mount table and symbolic links.

DJ

- Raw text -