Re: [aclug-L] ntmag

[Tom Hull <thull@xxxxxxxx>]

John,

Have you seen any good write-ups either exactly what has been done to
speed up SMP in 2.2 (possibly including reasons for not doing things --
I'm particularly curious about LWP's), and/or what is targeted for 2.3.

Read an interview with Linus in the premier issue of "Linux Magazine"
where he seems pretty gung-ho on beefing up SMP support.

John Goerzen wrote:
> 
> Here is a point-by-point response to the article.
> 
- [lots of good stuff]
>
> He then talks about a sendfile() API.  He states: "Without sendfile, a
> Web server. . . must first read the contents of the file into its
> memory".  This is FALSE.  Apache and similar servers use mmap(), which
> does not require them to read it into memory first.  Had he looked up
> the manpage for mmap, he would have seen this.  This makes the rest of
> his rant on this moot.

Sure, mmap() faults the file into memory, which is certainly faster
than read() would be. NT doesn't have mmap, so this point slips their
minds. However, Linux still has to write() the file back. NT sendfile()
does everything in supervisor mode, so it avoids the user mode mapping,
the context switches, and a bit of write overhead, and some cleanup.
So sendfile() should be faster than Linux, even if not as much faster
as the author seems to think. The savings is evidently enough for some
Unix vendors to implement the very similar sendv().

-- 
/*
 *  Tom Hull  ::  thull@xxxxxxxx  ::  http://www.ocston.org/~thull
 */