Re: [Zope3-Users] Re: Performance

2006-08-28 Thread Philipp von Weitershausen
David Johnson wrote:
 Okay. I think I figured it out. It seems that images and large objects are
 the main culprit.  We have very few images, mainly just icons for the
 content types and tabs.  
 
 I discovered this by using lynx and with IE disabling images.  
 
 So the new question is: how do I speed up image downloading? I'm guessing
 this has something to do with the ZODB?

Only if your images are in the ZODB...


___
Zope3-users mailing list
Zope3-users@zope.org
http://mail.zope.org/mailman/listinfo/zope3-users


Re: [Zope3-Users] Re: Performance

2006-08-28 Thread Tom Dossis
David Johnson wrote:
 Okay. I think I figured it out. It seems that images and large objects are
 the main culprit.  We have very few images, mainly just icons for the
 content types and tabs.  
 
 I discovered this by using lynx and with IE disabling images.  
 
 So the new question is: how do I speed up image downloading? I'm guessing
 this has something to do with the ZODB?

Hi David,

Here's some 'benchmarks' for you to compare against.

I accessed images both as resources and content objects.  Though there
was some difference b/w them it wasn't by any order of magnitude

I used a recent Zope3 version from svn.zope.org.  From the results it
looks like data bandwidth may be a factor to consider.

Also my benchmark client is written in python using twisted and it also
adds it's own (significant) overhead.

I also ran some tests against Zope instances and ZEO (results not
included), which didn't turn up any weird result.


object=neptune.jpg size=3.39MB


1/ Accessed as resource, content is served from the file system..

Requests:20
Errors:  0
Async level: 8
kBytes rcvd: 71097.0
Total time:  4.08 sec

Req rate:4.9 req/sec
Data rate:   17416.7 kB/sec
Get ave: 0.204 sec

 #Req  #Err Get(ave) (min)  (max)kB/s  URL
   20 0   1.458  0.855  1.695  2438.4
http://localhost:8082/@@/neptune.jpg


2/ Accessed as a Image content (ZODB) object with a Zope/ZODB instance..

Requests:20
Errors:  0
Async level: 8
kBytes rcvd: 71097.0
Total time:  4.09 sec

Req rate:4.9 req/sec
Data rate:   17397.5 kB/sec
Get ave: 0.204 sec

 #Req  #Err Get(ave) (min)  (max)kB/s  URL
   20 0   1.462  0.867  1.685  2431.2
http://localhost:8082/content/neptune.jpg



object=sts.jpg size=371kB


3/ Smaller image accessed as resource object

Requests:100
Errors:  0
Async level: 8
kBytes rcvd: 37996.2
Total time:  1.08 sec

Req rate:92.8 req/sec
Data rate:   35274.5 kB/sec
Get ave: 0.011 sec


 #Req  #Err Get(ave) (min)  (max)kB/s  URL
  100 0   0.085  0.035  0.109  4456.0  http://localhost:8082/@@/sts.jpg


4/ Smaller image accessed as content object

Requests:100
Errors:  0
Async level: 8
kBytes rcvd: 37996.2
Total time:  1.41 sec

Req rate:70.7 req/sec
Data rate:   26870.1 kB/sec
Get ave: 0.014 sec

 #Req  #Err Get(ave) (min)  (max)kB/s  URL
  100 0   0.112  0.042  0.156  3381.7
http://localhost:8082/content/sts.jpg

___
Zope3-users mailing list
Zope3-users@zope.org
http://mail.zope.org/mailman/listinfo/zope3-users


[Zope3-Users] Re: Performance

2006-08-27 Thread Philipp von Weitershausen
David Johnson wrote:
 Good morning. I’ve built and application for Zope 3 and it seems to run
 fairly slow (the logs show 2-5 seconds for a single page load).  The
 graphic usage is relatively low (I’ve tested the same html file in PHP
 and the speed is 0.10-0.20 seconds).

I'm not sure what graphics usage and testing the same html file in
PHP are supposed to mean. I also don't understand what such a test is
supposed to show.

There can be numerous bottlenecks in a Zope application. Rendering Page
Templates (which admittedly are not the fastest templating mechanism in
the world) is only one. Having unit test and integration tests for your
components can make it a lot easier to track down such bottlenecks. The
test runner also has a profiling option (--profile) that I might help
you show where you spend most of your time. It's a good idea to combine
this with --repeat=N so that the tests are run several times in a row
and will outweigh the overhead spent in the test runner in the statistics.

Philipp

___
Zope3-users mailing list
Zope3-users@zope.org
http://mail.zope.org/mailman/listinfo/zope3-users