On 3/22/2010 5:08 AM, Geoff Millikan wrote:
>> if your server averages 300 simulatious connections, you need
>> to start with 300 servers, and you never want it to drop
>> below that number.
> 
> Your experience might show otherwise however based on our experience - if we
> averaged 300 new customers/min at once (not 300 requests/sec) a
> MinSpareServers of 300 wouldn't be enough.

A connection is not a customer, as we discuss below.  I'm using the technical,
networking term connection, and this is what you can observe in mod_status
over a period of time to calculate an average (and min/max).

>> Yes, a browser can make multiple connections, but this is 
>> typically only two
>> parallel pipelines, perhaps even four.  
> 
> The "average" browser now makes 6 parallel connections per hostname per:
> http://www.browserscope.org/

Interesting research, thanks!

>> But 30 workers are not handling the 30 requests comprising
>> one user connecting to your site!!!  You just happened
>> to hit a magic correlation in your testing :)
> 
> I agree, the way I understand the prefork model to work, the 30 processes
> aren't each serving one of the 30+ requests this Customer's browser made (is
> keepalive tracked across processes?).  However, the way the testing worked
> out for us, it seems that way.  We did a lot of testing to come up with our
> numbers and just "ball parking" it, the number of servers seemed to work out
> best when matched to the number of requests per page.  

Fair enough, but if your testing was of ~30 requests, and we are believing that
the typical browser is making 6 simultaneous connections, then it sounds like
the real magic was 6 * fudge factor of 5 ;-P


---------------------------------------------------------------------
The official User-To-User support forum of the Apache HTTP Server Project.
See <URL:http://httpd.apache.org/userslist.html> for more info.
To unsubscribe, e-mail: users-unsubscr...@httpd.apache.org
   "   from the digest: users-digest-unsubscr...@httpd.apache.org
For additional commands, e-mail: users-h...@httpd.apache.org

Reply via email to