🚀 go-pugleaf

ken@yorku.ca (Kenneth Kafieh) writes:

>I had to resort to drastic mesures inorder to put this message here.

>I found that I was able to get 3-4 time faster performance by using good 
>old ANSI C file accessing routines (like  fgets, fseek, ftell, ...etc.)
>instead of C++ streams (like ifstream::getline, fstream::seekg, ...etc.)
>I wrote the program originally in C++ (a simple little phonebook for my
>own use) and it took about 9 seconds to scan a 100Kb file for a 3 byte
>string!!!  Fed up, I rewrote it in C and voila! Now it takes 2 secs!
>I'm using Borland C++ v2.0.

>Does that sound right?  Maybe my code was inefficiently written?
>I know C++ is generally slower than C but I didn't think that 
>file stuff would be too.   What did I do wrong? Am I missing
>something.  Maybe I'll post the code too, later. Has anyone
>else noticed this too?

I think that is about right. C++ file routines can be a little slow...

If the record length is more than the length of the data you are 
searching for, I would suggest putting a sorted index on the file. 

I have a fortune cookie program I wrote that currently is 1.4Mb long and 
has 10,000+ fortunes. I implemented a index on it, and made a file about 
41k long. It took about 5 minutes to create the index from scratch, but 
much less than a second to pick any particular fortune. Eg almost as soon 
as my finger clears the Enter key it prints a fortune.

Use a binary search on the index, get a pointer to the data file, and 
bingo you will get the record in question.

Otherwise; Sort the data and use a binary search on it.

Carl
-- 
>>>>> All flames sent to Demon Dragon familiar for appropriate response <<<<<
cwyles@nyx10.cs.du.edu | My words are the amalgamation of my experiences only
Good advice is ignored. Ok advice is used. Bad advice is flamed forever.