Re: sf squid branch

On Fri, Mar 18, 2005, Henrik Nordstrom wrote:
> On Fri, 18 Mar 2005, Adrian Chadd wrote:
>
> >I'm thinking of some simple feedback from the client -> server saying
> >"yup, I've copied a little more".
>
> Ok. Along the same lines I outlined some day ago then.

I'm sure its not a new concept - I'm just going to try implementing it. :)

> >The http.c side stuff can then stop reading when it finds itself too far
> >ahead, and resume reading when the clients have caught up.
>
> Another approach is to get rid of the concept of "ahead" entirely in this
> equation, and only have the server side read if there is clients looking
> for more data.. This should fit rather nicely with edge driven event
> notification systems and optimistic I/O operations (try first, wait for
> event later).

Trouble is, and its partly why I started the code changes I did pre
the C++ changeover, the store reads don't always have to be at the tail
end of the object. its not an insurmountable problem though.
thing is, it may be a bit intrusive into squid-2.5 and instead we may
wish to call it squid-2.6..

adrian

-- 
Adrian Chadd			"To believe with certainty we must first
<adrian@creative.net.au>	    begin by doubting."