[ECOS] Huge performance difference between read() and fread()

[Rutger Hofman]

Rutger Hofman wrote:
> Hi list,
> 
> I have a board with an PXA270 and 32MB of flash, part of which is 
> configured as a jffs2 flash file system with Flash/v2 and legacy 
> interface. I also have language/C and many other C components configured 
> in, the relevant functions are open()/read() and fopen()/fread().
> 
> Now, if I read a file of size 503549 on this jffs2 file system, the 
> performance is *wildly* different for read() and fread():
> 
> file "/fpga/guardian_v3_rev_b.svf" size 503549
> read file /fpga/guardian_v3_rev_b.svf size 503549 takes 0.863116
> fread file /fpga/guardian_v3_rev_b.svf size 503549 takes 16.601978
> 
> Times are in seconds.
> 
> Question: why are the timings a factor of 20 apart? Why are the timings 
> different at all? I'd think that both do more or less exactly the same 
> thing, although at a different interface.

This provides at least part of the answer:

the data above were obtained with a buffer size of 8K. It appears that 
FILEs have a buffer size of 256. If I change the buffer size in my test 
program to 256, I get the following timings:

file "/fpga/guardian_v3_rev_b.svf" size 503549
read file /fpga/guardian_v3_rev_b.svf size 503549 takes 15.767474
fread file /fpga/guardian_v3_rev_b.svf size 503549 takes 16.624629

OK, so the difference all but disappears. So if I want to have some 
performance in reading I'd better increase the FILE buffer size. I'll 
check for setvbuf and friends.

Rutger Hofman
VU Amsterdam


-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss