NetNews Offline 2/NetNews Offline Volume 2.iso/news/comp/lang/c-part1/3180

Internet Message Format | 1996-08-05 | 1.7 KB

Path: news.eclipse.net!usenet
From: [e-mail]
Newsgroups: comp.lang.c
Subject: Re: performance of fread
Date: 26 Jan 1996 17:08:42 GMT
Message-ID: <4eb1qq$[e-mail]>
References: <4eaecp$[e-mail]>
Reply-To: [e-mail]
NNTP-Posting-Host: so_69.eclipse.net
X-Newsreader: IBM NewsReader/2 v1.2

>My question: How time consuming is the overhead of calling fread reading from
>a large buffer compared to a for( ) loop which takes his data from a memory
>block which was filled by a single call of fread.

Well.  I'll tell you.  If this helps.  I had the same problem.  I had to read in an index
to a database ( C type structures ), that had over 1000 elements to the index.
Performing 1000 freads would take way to long, so we swapped out the un-used
section of memory at the time into XMS, freeing up about 50k of DOS memory in
the app, allocated the 40k needed to hold the data, and performed one big fread
into the 40k heap, then, memcpy'd the data into the structure pointer (already allo-
cated for the 40k), free'd up the 40k buffer, closed the file and reloaded the data
from XMS into it's right locations.

The optimized process takes about 2 seconds less than the origional code, and the
profiler shows an almost 25% reduction in code execution for that function, freeing
up much needed processor time.

The main problem here, is when we attempt to load the 120k record base into xms,
that damm near takes forever, so we parse load it, 20k initial, when they get 1/2 way
through the data view, we quickly load another 20k, you notice this on the pgu-pgd
combo's, but just regular scrolling or searching is fine.

Hope that helps.