Irrelevant ...mmap() VS read()

[Michael Widenius]

>>>>> "Mark" == Mark Papadakis <markp@stripped> writes:

Mark> Hi!
Mark> I am sorry if its kinda irrelevant but I wanted an opinion in this before
> actually doing it.

Mark> About 80.000 reads are perfomed in my application in about 20cs.

Mark> Would it be faster to dump open()/read() and use mmap() and access the abstract
> memory layer instead?
Mark> And if so, how much faster would it be?

Mark> One element at a time is being read ( i.e 1 integer ).

Mark> Thank you,

Hi!

On many OS reads are made by doing a mmap()+ a copy from memory + an
un-mmap().  (On Solaris someone claimed that each reads make 3 mmap() calls).

This means that using mmap directly will be faster, possible 0-20 %
faster...

Regards,
Monty