RE: [squid-users] squid 2 redirects but does not cache

> -----Original Message-----
> From: christophe.gravier@univ-st-etienne.fr
> [mailto:christophe.gravier@univ-st-etienne.fr]
> Sent: Monday, January 23, 2006 1:47 PM
> To: squid-users@squid-cache.org
> Subject: RE: [squid-users] squid 2 redirects but does not cache
> Importance: High
>
>
> >> -----Original Message-----
> >> From: christophe.gravier@univ-st-etienne.fr
> >> [mailto:christophe.gravier@univ-st-etienne.fr]
> >> Sent: Monday, January 23, 2006 11:22 AM
> >> To: Chris Robertson
> >> Subject: RE: [squid-users] squid 2 redirects but does not cache
> >> >> -----Original Message-----
> >> >> From: Christophe Gravier
> >> >> [mailto:christophe.gravier@univ-st-etienne.fr]
> >> >> Sent: Sunday, January 22, 2006 5:26 AM
> >> >> To: squid-users@squid-cache.org
> >> >> Subject: [squid-users] squid 2 redirects but does not cache
> >> >>
> >> >>
> >>
> >> In fact, I wanted to run it as accelerator (with caching). In
> >> squid.conf I
> >> read:
> >> # NOTE: enabling httpd_accel_host disables proxy-caching and #
> >> ICP. If you want these features enabled also, set
> >> # the 'httpd_accel_with_proxy' option.
> >>
> >> So that's what I tried.
> >>
> >
> > So it's a definition problem. It is a bit ambiguous. Generally, an
> > accelerator accepts requests from the internet and caches
> > pages from a
> > small number of servers that it is "responsible" for. A
> > proxy accepts
> > requests from the small number of clients it is "responsible" and
> > caches pages from the internet. Having the option enabled shouldn't
> > hurt anything, but I think it's unnecessary.
> >
>
> So "accelerator" means reverse proxy. (I known understand your
> astonishment upon this part of the configuration !)
>
> >>
> >> Well, I am aware of cookies (it's embedded in my plone solution).
> >> Regarding caching itself, I don't understand why under heavy load
> >> (simulated with ab apache benchmark) they are really bad
> >> performances and
> >> meanwhile I see in store.log ONLY "RELEASE" operation.
> >
> > I would imagine that's fairly normal behavior for an
> > accelerator. You
> > should only see a CREATE action when a new object is
> > encountered that
> > can be cached, but isn't. S
>
> Sorry, that's what I don't understand. (my fault ;)).
> If it is responsible for one server for example, why, even upon
> restarting, didn't I see any trace of CREATE for object
> cached (the object
> the reverse proxy caches for "relieving" the servers ?)

Restarting the proxy will not clear out the cache. Restarting the webserver will not change the content. No CREATE actions will occur without new (non-cached) objects.

>
> > WAPOUT means that the object was in memory
> > and was written to disk. SWAPIN is just the opposite. I think I
> > remember reading that there is a problem with the 2.5
> > branch of Squid
> > which prevents it from performing a SWAPIN (which isn't as big a
> > problem as it sounds. Most operating systems cache file accesses to
> > memory), but I might be mistaken. Watch the unfiltered access log.
> > You should see some TCP_HITs and TCP_MEM_HITs. As for the poor
> > performance, have you had a look at
> > http://plone.org/documentation/tutorial/optimizing-plone/?
> > The page at
> >
> > http://plone.org/documentation/tutorial/optimizing-plone/what-to-cache
> > seems to have some hints at making plone play well with caches.
> >
>
> Ok I got it.
>
> Regarding plone configuration, I just follow limi's
> recommandation. The
> fact is in his example he received :Requests per second:
> 864.07 [#/sec]
> (mean)
> I do receive
> Requests per second: 3 [#/sec] (mean)
>
> Basically, "pure" plone without *any* cahcing policy is meant
> to response
> to 0,7 request per sec.
> Using wisely RamCache and HttpCache plone product raise it to
> 3 request
> per sec. (it's so ok for a single use accessing the website to see it
> being fast).
> Now it needs scalability (reverse proxy) to be able to
> maintain service
> under loads.
> I do think squid will change this rate significantly. (in fact with or
> without, i don't gain any digits).
>
> Making tail -f on access.log,
>
> If I launch apache benchmark on one side, and tail -f
> access.log the other
> side, I *only* have those lines:1138052704.779 401 161.3.50.16
> TCP_MISS/20024213 GET
> http://161.3.50.16:81/VirtualHostBase/http/ist-guizay.univ-st-
> etienne.fr:80/VirtualHostRoot/Portail- DIRECT/161.3.50.16
> text/html1138052705.231 450 161.3.50.16
> TCP_MISS/200 24213
> GEThttp://161.3.50.16:81/VirtualHostBase/http/ist-guizay.univ-
> st-etienne.fr:80/VirtualHostRoot/Portail- DIRECT/161.3.50.16
> text/html1138052705.560 327 161.3.50.16

SNIP

>
> As if it only tries to match against cache only for text/html and not
> images etc. ...

Well, one thing I am seeing is...

GET /Portail HTTP/1.1
...

HTTP/1.x 200 OK
Date: Mon, 23 Jan 2006 23:27:29 GMT
...
Expires: Sat, 1 Jan 2000 00:00:00 GMT
...

...the main Plone page is serving expired content. Figure out why that is, and I'm sure your benchmark results will rise.

Chris
Received on Mon Jan 23 2006 - 16:45:02 MST