Reading Large Files in C

To read an entire file efficiently you need to know the file size FIRST.

So whats the fastest way to programatically find the file size. If you read stackoverflow you’ll find a number of suggestions, but many use fseek to jump to the end of the file, find your location and use that as the file size…

This works, but is 30% slower than using stat or fstat

I just ran a bunch of tests comparing using seek, lseek, stat, and fstat also comparing using file streams and file descriptors to see what seems to be the fastest. For the test I create a 100M file.

TL;DR — using file descriptors, fstat and read was the fastest and using file streams and seek was the slowest. Go to the bottom to see the real slowest.

For the test I ran this on a small Linux box I have running a headless ArchLinux server. I ran the test: checking the file size, malloc a buffer, read the entire file into the buffer, close the file, free the buffer.

I ran the test 3 times with 1000 cycles each time and using clock_gettime to calculate the elapsed time.

Just simply comparing JUST the time it takes to get the file size using stat or fstat were at least 30% faster than using seek or lseek.

Comparing just the speed of using file streams vs file descriptors, they were pretty nearly the same — descriptors were about 1–3% faster.

In comparing getting the file size, opening the file, malloc a buffer, read the entire 100M, close the file and free the buffer -- using file descriptors and fstat were 6-8% faster than using seek or lseek. Probably because the bulk of the time is spent in the file read vs the getting the file size, which dilutes the overall performance benefit.

BTW — do not use fgetc and read the file 1 character at a time. This is crazy inefficient and really really slow! Like 1700% slower!!!!