httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Brian Havard" <bri...@kheldar.apana.org.au>
Subject Re: cvs commit: apache-2.0/src/lib/apr/mmap/unix mmap.c
Date Fri, 07 Apr 2000 12:15:26 GMT
On Fri, 7 Apr 2000 00:34:33 -0400 (EDT), rbb@apache.org wrote:

>> For OS/2 I implemented buffered I/O entirely within APR, not using FILE *.
>> The same approach would work on other platforms. You could even just use
>> the same code & switch the OS/2 API calls to whatever's best on each
>> platform.
>> 
>> If you like I can do this for unix & Win32. Note that the code I used was
>> derived from a C++ class I've been using in my own projects for several
>> years so it should be fairly solid (though I could have broken something in
>> the C++->C conversion...).
>
>The general feeling was that Apache is doing the buffering with the BUFF
>code, and we don't need to do buffering twice in Apache (once in Apache
>and once in APR).

Yes, BUFF is used for files being served but ap_fgets is used for things
like reading config files & without buffering you end up with a read()
syscall or the platform equivalent for every individual byte. This may or
may not have a noticable impact on performance in Apache (mod_auth may),
you'd have to do some benchmarking to find out.

-- 
 ______________________________________________________________________________
 |  Brian Havard                 |  "He is not the messiah!                   |
 |  brianh@kheldar.apana.org.au  |  He's a very naughty boy!" - Life of Brian |
 ------------------------------------------------------------------------------


Mime
View raw message