On Tue, 26 Sep 2006, Graham Leggett wrote:
Niklas Edmundsson wrote:
* Realising that a file is a file and can be copied as such, without
reading the whole thing into memory first.
* When a file is cached by copying, replace the brigade with a new one
refering to the cached file so we don't have to read the file from
the backend again when sending a response to the client.
As I read the code, the copy is completed before an attempt is made to
deliver the copy to the network. This should in theory stop a slow initial
client from holding up faster following clients, if the caching is still "in
transit".
Is this correct?
This is the original design of mod_disk_cache (which isn't changed by
this patch), so yes.
In practice this isn't enough when dealing with large files, so in our
production code (the hideously large jumbopatch) this is fixed by
read-while-caching and spawning a thread to do the caching in the
background while delivering the response (by read-while-caching) to
the client that initiated the caching.
/Nikke
--
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Niklas Edmundsson, Admin @ {acc,hpc2n}.umu.se | [EMAIL PROTECTED]
---------------------------------------------------------------------------
"Why don't you go teach someone who actually needs to learn?!" - Buffy
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
