Path: utzoo!utgpu!news-server.csri.toronto.edu!bonnie.concordia.ca!thunder.mcrcim.mcgill.edu!snorkelwacker.mit.edu!shelby!agate!usenet.ins.cwru.edu!gatech!uflorida!reef.cis.ufl.edu!jdb
From: jdb@reef.cis.ufl.edu (Brian K. W. Hook)
Newsgroups: comp.os.msdos.programmer
Subject: Re: Fastest dump to disk?
Message-ID: <26994@uflorida.cis.ufl.EDU>
Date: 18 Feb 91 13:56:08 GMT
References: <91043.105618JJLUCSY@MTUS5.BITNET> <2184@umriscc.isc.umr.edu> <1091@usage.csd.unsw.oz.au>
Sender: news@uflorida.cis.ufl.EDU
Organization: UF CIS Dept.
Lines: 46

In article <1091@usage.csd.unsw.oz.au> pwb@newt.phys.unsw.OZ.AU (Paul W. Brooks) writes:
|>In article <2184@umriscc.isc.umr.edu>, mcastle@mcs213f.cs.umr.edu (Mike Castle {Nexus}) writes:
|>> In article <91043.105618JJLUCSY@MTUS5.BITNET> JJLUCSY@MTUS5.BITNET writes:
|>> >Could someone please tell me what is the fastest way to dump a chunk of memory
|>> >to disk using either Turbo C v.2 or Turbo C++. I'm not to particular on the
|>> >method, but it should be real quick. Thanks in advance.
|>> >
|>> 
|>> Try fwrite.  The prototype is:
|>> 
|>> size_t fwrite(const void *ptr, size_t size, size_t n, FILE *stream);
|>> 
|>> It starts at *ptr and writes n records of size size to file pointed to by 
|>> steam.
|>
|>Sorry - incorrect. The fastest way is using low-level file access -
|>write(). Using fwrite() requires the disk library internals to keep up
|>to date with internal buffers, etc., and is quite slow. In my systems I
|>usually get at least 3 times the speed when using write() instaed of
|>fwrite()!
|>	There are a couple of things to watch, though. Preferably the
|>file handle is obtained through the low-level call to open() or creat(). If you
|>need to write to a file opened using the stream method (fopen()), and
|>get to the underlying file handle using fileno(stream), then beware that
|>the buffers probably will not be correct, and any later streams-based
|>calls (fwrite(), fread(), fgetc(), etc etc) may fail unpredictably.
|>There are techniques to bring the low-level and higher-level routines
|>back in line, e-mail me if you need some hints. In my software I have to
|>manipulate 500K astronomical images, and use only the low-level file
|>operations:
|>
|>Anyone know of any faster way of accessing large amounts of disk data?
|>I'd really appreciate hearing it. :-)
|>Regards,

I have heard somewhere that open(), write(), close(), and read() were not
defined in the ANSI standard because of their reliance on handles (  a DOS
and UNIX system specific thing).  Thus fopen() and the other stream based
functions would seem to be a bit more portable.

However, speed seems to be the issue here, so I would also look in the TC++
manual under _open() and the other file functions with a leading
underscore.  I don't know if they are any faster, but they seem to a bit
more low level.  Inline asm might be also used...:)

Brian