Re: Web simulation (fwd)

[Kaichuan He <he@expert.cc.purdue.edu>]

Haobo:
	Thanks for the explanation. Now I understands: the centralized
page pool serve 2 different purpose for server and client,
	for server: it provide page information, just like a repository
	for client: it provide what pages are in the server, and also
		    provide a "URL request sequence" generator by
		    random number generator or by a trace file.
	
	For one client+one cache+one server model, this is OK. 
But for multi clients+mutli client cache+one server model, will
this also work?  I mean, now, if we still use only one page-pool,
then how do we distinguish page attribute like "hit-count/ref count"
for different page?  Or can I say that the number of page-pool
needed is equal with the number of caches simulated?


Thanks,
 


Kaichuan


On Mon, 17 May 1999, Haobo Yu wrote:

> In order for the client to generate a request, it must which page a server
> has. Also, when the server starts up, it needs to know what are the
> attributes of all of its pages. That page pool serves this purpose. Its
> name should have been something like "Page generator" to avoid the
> confusion. 
> 
> A server needs to bind to a page generator in order to initialize its page
> information. For example, if one uses a proxy trace to drive a simulation,
> and the server needs to load the page information from the page generator
> at the beginning. It is less desirable to do this let every client and
> every server to load this trace file individually, and a centralized
> page pool allows all clients and servers to share this information.
> 
> Hope this helps.
>