httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dean Gaudet <dgau...@arctic.org>
Subject Re: Apache 2.0 ideas
Date Fri, 06 Nov 1998 16:03:55 GMT


On Tue, 3 Nov 1998, Mark D. Anderson wrote:

> On a related note, if there is a choice of methods (e.g. sendfile
> vs. memory), could the web server just try them both on successive
> requests and in the future (on that resource), pick the faster?
> There would have to be adjustment for hotness of the caches, but
> you get the idea.
> 
> That would relieve people of tuning their system manually.

But we can never relieve people of tuning their system manually.  Suppose
the file being served is a 100Mb movie file, and the system serving it has
only 64Mb (silly mistake, web servers should have gobs of RAM, but I
digress).  It could be that at 4am the first time this is requested
there's no other processes competing for memory and the server figures out
that mmap() is the "fastest" way to serve this file.  But at 12 noon when
there's competition for memory, mmap() could completely suck, and
read()/write() may be the way to go.  Who knows.

Get the common case "good enough" and leave performance tuning for the few
percent that need it.  But give them the tools and documentation they need
to make their choices. 

Dean


Mime
View raw message