Requesting real-world stats, please

[Cosimo Streppone]

On Tue, 25 May 2010 14:28:41 +0200, Poul-Henning Kamp <phk at phk.freebsd.dk>  
wrote:

>
> I may have "misunderestimated" the dynamics of how objects live in
> Varnish caches and thus taken a wrong optimization decision with
> the critbit hasher.
>
> I am looking for data to figure this out, specifically the values of
> the two varnishstat counters:
>
> 	"N struct objecthead."
> and
> 	"N struct object"
>
> Along with information about which version of varnish, and which, if
> any -h argument you use.

I don't have any critbit info, since we're running on 2.0.4,

   ii  libvarnish1  2.0.4-5~bpo50+1  shared libraries for Varnish
   ii  varnish      2.0.4-5~bpo50+1  a state-of-the-art, high-performance  
HTTP accelerator

Here's the counters from one installation, hash default settings
serving ~100 req/s of very small objects, 2 twin machines,

   n_object               100585          .   N struct object
   n_objecthead           103700          .   N struct objecthead

   n_object               102270          .   N struct object
   n_objecthead           102294          .   N struct objecthead

Maybe I should use -h with a higher number? :)

This other here is a different installation, ~500-1000 req/s
of medium/big files/pics and longer ttl (10 days default),
again 2 twin machines,

   n_object               303979          .   N struct object
   n_objecthead           299199          .   N struct objecthead

   n_object               506522          .   N struct object
   n_objecthead           504143          .   N struct objecthead

Here I'm using '-h classic,500009'

-- 
Cosimo