[thelist] Spreading sites between different hosts

[Anthony Baratta]

Peter Small wrote:
> 
> Thanks Anthony for your response.
> 
> The implications of this are very interesting. If I may, I'd like to extend
> this thought to the components of a Web page.
> 
> (1)
> Ignoring the different speeds of response between different servers through
> traffic etc,would there be a time penalty if a page were designed to have
> ten gifs with each of the gifs located on a different servers (i.e. ten
> gifs and eleven servers)?

Remember that a web browser can only communicate so many different streams at once. I
see a definite lag on the large commercial sites when the ads (which are being served
by multiple ad servers) are loading. I think Netscape only uses 4-5 streams. So with
10 servers to communicate with to build one page, I would think that this 'dispersed'
page would be slower than communicating with one single page. Especially since the
client would have the choke point on bandwidth - don't forget broadband is not a
reality for 95% of the world.
 
> (2)
> Would there be a time penalty if items included in a Web page were accessed
> from several different databases on different servers from that holding the
> Web page (assuming all the databases were equally fast)

Depends who's doing the communicating. If the target web server is calling all the
other databases, I would think there is a slight performance hit to build up those
separate DB connections (ther are not "cheap" to perform) - but if you were doing
several large CPU intensive queries, using several servers with one query each
"should" be faster than one server with all the queries. But this assumes that you
can do all the queries "at once". Most programs are serial, not asyncronous. Which
means that using different servers would not be any an faster since you have to
communicate with them one at a time, gaining none of the parallel query benefits.

If the browser is doing the communicating we are back to the client limitation in
question one.
 
> (3) If a cgi script updated the visitor counter on a Web page, would there
> be a time penalty if that cgi script was on another server?
> 

Many free counters and web stats programs use this method. Theoretically, there
should be no penalty, except for the slight hit the first time to resolve the other
server URL. DoubleClick and all the other ad companies use this method, so the
performance hit on average can't be that bad. ;-)

-- 
Anthony Baratta
President
KeyBoard Jockeys
                    South Park Speaks Version 3 is here!!!
                       http://www.baratta.com/southpark
                              Powered by Tsunami