Metadata in zope / binary data on FS, was Re: [ZODB-Dev] __del__ with Persistent objects

[Jeremy Hylton]

On Fri, 2003-07-18 at 13:06, Paul Winkler wrote:
> On Fri, Jul 18, 2003 at 06:01:57PM +0100, Toby Dickenson wrote:
> > Alternatively, squid may be able to help in another way. You could make your 
> > files cacheable with "Cache-Control:public,s-maxage=0,proxy-revalidate".... 
> > That allows shared caches to store the file, but *always* check authorization 
> > with the Zope origin server. On the second and subsequent request, the file 
> > body will be served by squid.
> 
> Now *that* is a cool trick.
> Don't know if it helps in my case though... first i'd have to get our
> sysadmin to agree to add squid to our already-complex system.
> and caching might not help that much: typical usage for us is that each 
> file is rarely accessed but when the system is busy, several
> different files are sometimes requested at the same time and the server
> crawls.

Chris and Shane had some ideas about how to improve ZServer for large
files IIRC.  I believe the problem is that one of Zope's threads get
used for the large file for the entire time the client is downloading. 
That makes Zope slow for everything else.

Jeremy