pramodh mallipatna wrote:
> 2. Input and Output files mapped to the user space (using mmap) and just
> mem-copying from input to the output file. - According to an example in
> Richard Stevens "Advanced Programming in the UNIX environment"
>
> 3. My own code which basically does the transfer at the buffer cache level
> (I read a buffer using generic_file_read() and use ext2_file_write() to
> write it)
>
>
> I find case 2 better than the rest (because of which I am slightly
> confused) and case 3 better than case 1 (which is expected).
>
> My question is:
> ---------------
> Why is case 3 not on par with case 2?
>
> I understand that in case 3, I read a page at a time (as generic_file_read
> reads 4 BLOCKSIZE'd chunks into a page and ext2_file_write writes 1
> BLOCKSIZE at a time).
>
> But, mmap() and memcpy() should ultimately result in the same calls
> at the buffer cache level, right?
>
> So, how is it I am getting better throughput in the mmap case? Is there
> something more to it, like some caching effects etc.? How does file mmap
> work, in that case?
mmap() does more than access the buffer cache. (BTW 2.3.x kernels it
access the page cache instead but the results are similar).
With mmap() and memcpy(), each new page accessed implies the overhead of
a page fault and the mapping setup for that page. There is also some
CPU overhead in looking things up through multiple page mappings: the
TLB (CPU page mapping) cache is finite, and there is a small cost
associated as each entry is refilled.
-- Jamie
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sun Apr 23 2000 - 21:00:10 EST