You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@forrest.apache.org by Ross Gardler <rg...@apache.org> on 2005/10/05 12:21:21 UTC
Performance in views
From the FT IRC logs:
+Oct 05 15:00:15 <kronenpj> On my quip regarding cocoon caching... That
site took 50 minutes to build using 0.8 & views.
+Oct 05 15:00:21 <xley> Ah yeah.
+Oct 05 15:00:34 <kronenpj> With essentially the same content the pelt
site takes much less.
+Oct 05 15:00:35 <xley> good success story
+Oct 05 15:00:40 <kronenpj> :)
+Oct 05 15:01:12 <xley> hey this is my worry with current views
development ...
+Oct 05 15:01:37 <xley> i get a sneaking feeling that it doesn't use
cocoon ideally.
+Oct 05 15:01:58 <kronenpj> I think you're probably right.
One reason for this poor performance is that xinclude is used
extensively. These reqeusts cannot be cached. Therefore everything is
generated *every* time it is requested.
We need to switch this to cinclude.
Ross
Re: Performance in views
Posted by Ross Gardler <rg...@apache.org>.
Thorsten Scherler wrote:
...
>>
>>One reason for this poor performance is that xinclude is used
>>extensively.
>
>
> Actually there is "only" one part of the code depending on a include
> with a xpointer. That is the reason I used xinclude. AFAIK cinclude does
> not support xpointer, right?
I think you are correct - not checked though.
> I need to have a look how to rewrite this.
Can we write a custom generator? I know it is your medium-long term goal
to move stuff to Java. In my opinion this is not an urgent issue, as
long as we know where we are heading with it and that we can solve the
problem.
>>These reqeusts cannot be cached. Therefore everything is
>>generated *every* time it is requested.
>>
>>We need to switch this to cinclude.
Since that's a search and replace we may as well do it where we can.
Ross
Re: Performance in views
Posted by Thorsten Scherler <th...@apache.org>.
El jue, 06-10-2005 a las 18:34 +1000, David Crossley escribió:
...
> > Why do you think that David?
> > Can you please show us!
>
> I needed to do a better analysis before saying anything,
> but since you ask now ...
>
> Have a look at a views enabled site using the Cocoon Profiler:
> http://forrest.apache.org/docs_0_80/howto/howto-dev.html#debug
>
> There seem to be a number of components not being cached.
>
> As i said i would rather have done some anaylysis first.
> At this stage its like i said on IRC: "a feeling" only.
>
> Aside from my feelings, Paul's build times are frightening.
> One of us should views-enable our site-author and see
> what build time happens for us.
Will do that. Thx for the link. ;-)
salu2
--
thorsten
"Together we stand, divided we fall!"
Hey you (Pink Floyd)
Re: Performance in views
Posted by CFAS Webmaster <we...@cfas.org>.
David Crossley wrote:
>David Crossley wrote:
>
>
>>CFAS Webmaster wrote:
>>
>>
>>>Here's a second try...
>>>
>>>Tigris skin:
>>>Part 1: http://forrest.pastebin.com/385730
>>>Part 2: http://forrest.pastebin.com/385731
>>>
>>>
>>Did you grasp the importance of issue FOR-677 below?
>>You should fix that issue first. It is causing documents
>>to be duplicated.
>>
>>
I did, but I wanted a complete record to be documented first...
>
>Try looking through your site.xml for hrefs
>with leading slash, also skinconf.xml, also the
>links in xdocs if not using site: style links.
>
>
>
I'll do that. I know I have quite a few like that due to the way the
site was originally created that I never removed.
>-David
>
>
Thanks!
-Paul
Re: Performance in views
Posted by David Crossley <cr...@apache.org>.
David Crossley wrote:
> CFAS Webmaster wrote:
> > Here's a second try...
> >
> > Tigris skin:
> > Part 1: http://forrest.pastebin.com/385730
> > Part 2: http://forrest.pastebin.com/385731
>
> Did you grasp the importance of issue FOR-677 below?
> You should fix that issue first. It is causing documents
> to be duplicated.
Try looking through your site.xml for hrefs
with leading slash, also skinconf.xml, also the
links in xdocs if not using site: style links.
-David
Re: Performance in views
Posted by David Crossley <cr...@apache.org>.
CFAS Webmaster wrote:
> Here's a second try...
>
> Tigris skin:
> Part 1: http://forrest.pastebin.com/385730
> Part 2: http://forrest.pastebin.com/385731
Did you grasp the importance of issue FOR-677 below?
You should fix that issue first. It is causing documents
to be duplicated.
-David
> Views:
> Part 1: http://forrest.pastebin.com/385761
> Part 2: http://forrest.pastebin.com/385762
>
> However, my "feeling" that pages were being processed about 4 times
> appears not to be the case. I guess that means that there's something
> taking a lot longer to process with views than with tigris, or that
> there's a lot more stuff being generated...
>
> -Paul
>
> CFAS Webmaster wrote:
>
> >Sorry, I royally messed up on posting to pastebin... I'll post it
> >again and send the links...
> >
> >By the way, do we use http://forrest.pastebin.com or something like
> >it? It can give us a semi-private posting area outside of the main page.
> >
> >-Paul
> >
> >David Crossley wrote:
> >
> >>CFAS Webmaster wrote:
> >>
> >>
> >>>If anyone is interested, I have posted a complete log of the views &
> >>>pelt build so everyone can see how many times various pages are
> >>>compiled...
> >>>
> >>>Views: http://pastebin.com/385671 (**Huge**) & ~30 minutes
> >>>Pelt: http://pastebin.com/385673 ~4 minutes...
> >>>
> >>>While the sites aren't *exactly* the same, they are the same except
> >>>for the index page which was rearranged. All other content is
> >>>synchronized by SVN.
> >>
> >>I have only looked at the "pelt" build so far
> >>and i see that you suffer from a similar problem
> >>to what we had with the Forrest/site-author/ site.
> >>e.g.
> >>---------
> >>...
> >>798: * [238/166] [0/40] 0.301s 9.4Kb /about.html
> >>...
> >>1290: * [440/124] [0/40] 0.232s 9.4Kb about.html
> >>...
> >>---------
> >>
> >>http://issues.apache.org/jira/browse/FOR-677
> >>"leading slash in gathered URIs causes double the number of links to
> >>be processed"
> >>
> >>If you can fix that problem first, then the issues
> >>with "views" might be isolated to something else.
> >>
> >>Also, not that it matters with this, but you are not
> >>using the "pelt" skin but the old "tigris" instead.
> >>
> >>-David
Re: Performance in views
Posted by CFAS Webmaster <we...@cfas.org>.
Here's a second try...
Tigris skin:
Part 1: http://forrest.pastebin.com/385730
Part 2: http://forrest.pastebin.com/385731
Views:
Part 1: http://forrest.pastebin.com/385761
Part 2: http://forrest.pastebin.com/385762
However, my "feeling" that pages were being processed about 4 times
appears not to be the case. I guess that means that there's something
taking a lot longer to process with views than with tigris, or that
there's a lot more stuff being generated...
-Paul
CFAS Webmaster wrote:
> Sorry, I royally messed up on posting to pastebin... I'll post it
> again and send the links...
>
> By the way, do we use http://forrest.pastebin.com or something like
> it? It can give us a semi-private posting area outside of the main page.
>
> -Paul
>
> David Crossley wrote:
>
>> CFAS Webmaster wrote:
>>
>>
>>> If anyone is interested, I have posted a complete log of the views &
>>> pelt build so everyone can see how many times various pages are
>>> compiled...
>>>
>>> Views: http://pastebin.com/385671 (**Huge**) & ~30 minutes
>>> Pelt: http://pastebin.com/385673 ~4 minutes...
>>>
>>> While the sites aren't *exactly* the same, they are the same except
>>> for the index page which was rearranged. All other content is
>>> synchronized by SVN.
>>>
>>
>>
>> I have only looked at the "pelt" build so far
>> and i see that you suffer from a similar problem
>> to what we had with the Forrest/site-author/ site.
>> e.g.
>> ---------
>> ...
>> 798: * [238/166] [0/40] 0.301s 9.4Kb /about.html
>> ...
>> 1290: * [440/124] [0/40] 0.232s 9.4Kb about.html
>> ...
>> ---------
>>
>> http://issues.apache.org/jira/browse/FOR-677
>> "leading slash in gathered URIs causes double the number of links to
>> be processed"
>>
>> If you can fix that problem first, then the issues
>> with "views" might be isolated to something else.
>>
>> Also, not that it matters with this, but you are not
>> using the "pelt" skin but the old "tigris" instead.
>>
>> -David
>>
>>
Re: Performance in views
Posted by CFAS Webmaster <we...@cfas.org>.
Sorry, I royally messed up on posting to pastebin... I'll post it again
and send the links...
By the way, do we use http://forrest.pastebin.com or something like it?
It can give us a semi-private posting area outside of the main page.
-Paul
David Crossley wrote:
>CFAS Webmaster wrote:
>
>
>>If anyone is interested, I have posted a complete log of the views &
>>pelt build so everyone can see how many times various pages are compiled...
>>
>>Views: http://pastebin.com/385671 (**Huge**) & ~30 minutes
>>Pelt: http://pastebin.com/385673 ~4 minutes...
>>
>>While the sites aren't *exactly* the same, they are the same except for
>>the index page which was rearranged. All other content is synchronized
>>by SVN.
>>
>>
>
>I have only looked at the "pelt" build so far
>and i see that you suffer from a similar problem
>to what we had with the Forrest/site-author/ site.
>e.g.
>---------
>...
>798: * [238/166] [0/40] 0.301s 9.4Kb /about.html
>...
>1290: * [440/124] [0/40] 0.232s 9.4Kb about.html
>...
>---------
>
>http://issues.apache.org/jira/browse/FOR-677
>"leading slash in gathered URIs causes double the number of links to be processed"
>
>If you can fix that problem first, then the issues
>with "views" might be isolated to something else.
>
>Also, not that it matters with this, but you are not
>using the "pelt" skin but the old "tigris" instead.
>
>-David
>
>
Re: Performance in views
Posted by David Crossley <cr...@apache.org>.
CFAS Webmaster wrote:
> >
> If anyone is interested, I have posted a complete log of the views &
> pelt build so everyone can see how many times various pages are compiled...
>
> Views: http://pastebin.com/385671 (**Huge**) & ~30 minutes
> Pelt: http://pastebin.com/385673 ~4 minutes...
>
> While the sites aren't *exactly* the same, they are the same except for
> the index page which was rearranged. All other content is synchronized
> by SVN.
I have only looked at the "pelt" build so far
and i see that you suffer from a similar problem
to what we had with the Forrest/site-author/ site.
e.g.
---------
...
798: * [238/166] [0/40] 0.301s 9.4Kb /about.html
...
1290: * [440/124] [0/40] 0.232s 9.4Kb about.html
...
---------
http://issues.apache.org/jira/browse/FOR-677
"leading slash in gathered URIs causes double the number of links to be processed"
If you can fix that problem first, then the issues
with "views" might be isolated to something else.
Also, not that it matters with this, but you are not
using the "pelt" skin but the old "tigris" instead.
-David
Re: Performance in views
Posted by CFAS Webmaster <we...@cfas.org>.
David Crossley wrote:
>Thorsten Scherler wrote:
>
>
>>Ross Gardler escribi??:
>>
>>
>>> From the FT IRC logs:
>>>
>>>+Oct 05 15:00:15 <kronenpj> On my quip regarding cocoon caching... That
>>>site took 50 minutes to build using 0.8 & views.
>>>+Oct 05 15:00:21 <xley> Ah yeah.
>>>+Oct 05 15:00:34 <kronenpj> With essentially the same content the pelt
>>>site takes much less.
>>>+Oct 05 15:00:35 <xley> good success story
>>>+Oct 05 15:00:40 <kronenpj> :)
>>>+Oct 05 15:01:12 <xley> hey this is my worry with current views
>>>development ...
>>>+Oct 05 15:01:37 <xley> i get a sneaking feeling that it doesn't use
>>>cocoon ideally.
>>>
>>>
>>Why do you think that David?
>>Can you please show us!
>>
>>
>...
>Aside from my feelings, Paul's build times are frightening.
>One of us should views-enable our site-author and see
>what build time happens for us.
>
>-David
>
>
If anyone is interested, I have posted a complete log of the views &
pelt build so everyone can see how many times various pages are compiled...
Views: http://pastebin.com/385671 (**Huge**) & ~30 minutes
Pelt: http://pastebin.com/385673 ~4 minutes...
While the sites aren't *exactly* the same, they are the same except for
the index page which was rearranged. All other content is synchronized
by SVN.
Thanks!
-Paul
Re: Performance in views
Posted by David Crossley <cr...@apache.org>.
Thorsten Scherler wrote:
> Ross Gardler escribi??:
> > From the FT IRC logs:
> >
> > +Oct 05 15:00:15 <kronenpj> On my quip regarding cocoon caching... That
> > site took 50 minutes to build using 0.8 & views.
> > +Oct 05 15:00:21 <xley> Ah yeah.
> > +Oct 05 15:00:34 <kronenpj> With essentially the same content the pelt
> > site takes much less.
> > +Oct 05 15:00:35 <xley> good success story
> > +Oct 05 15:00:40 <kronenpj> :)
> > +Oct 05 15:01:12 <xley> hey this is my worry with current views
> > development ...
> > +Oct 05 15:01:37 <xley> i get a sneaking feeling that it doesn't use
> > cocoon ideally.
>
> Why do you think that David?
> Can you please show us!
I needed to do a better analysis before saying anything,
but since you ask now ...
Have a look at a views enabled site using the Cocoon Profiler:
http://forrest.apache.org/docs_0_80/howto/howto-dev.html#debug
There seem to be a number of components not being cached.
As i said i would rather have done some anaylysis first.
At this stage its like i said on IRC: "a feeling" only.
Aside from my feelings, Paul's build times are frightening.
One of us should views-enable our site-author and see
what build time happens for us.
-David
Re: Performance in views
Posted by Thorsten Scherler <th...@apache.org>.
El mié, 05-10-2005 a las 11:21 +0100, Ross Gardler escribió:
> From the FT IRC logs:
>
> +Oct 05 15:00:15 <kronenpj> On my quip regarding cocoon caching... That
> site took 50 minutes to build using 0.8 & views.
> +Oct 05 15:00:21 <xley> Ah yeah.
> +Oct 05 15:00:34 <kronenpj> With essentially the same content the pelt
> site takes much less.
> +Oct 05 15:00:35 <xley> good success story
> +Oct 05 15:00:40 <kronenpj> :)
> +Oct 05 15:01:12 <xley> hey this is my worry with current views
> development ...
> +Oct 05 15:01:37 <xley> i get a sneaking feeling that it doesn't use
> cocoon ideally.
Why do you think that David?
Can you please show us!
> +Oct 05 15:01:58 <kronenpj> I think you're probably right.
>
> One reason for this poor performance is that xinclude is used
> extensively.
Actually there is "only" one part of the code depending on a include
with a xpointer. That is the reason I used xinclude. AFAIK cinclude does
not support xpointer, right?
I need to have a look how to rewrite this.
> These reqeusts cannot be cached. Therefore everything is
> generated *every* time it is requested.
>
> We need to switch this to cinclude.
See above does cinclude support xpointer, if so then we can switch right
away.
salu2
--
thorsten
"Together we stand, divided we fall!"
Hey you (Pink Floyd)