A simple benchmark getting the site object


Ho many of you complain about Plone performances? The latest Plone release are said to be faster than the former ones.

Her is a sample benchmark that compares the three ways to get the portal (Plone) object.

  • The first one, I found it when wandering in the Zope 3 subset that ships with Zope 2.10.
  • The second one is the recommanded practice since Plone 3, getting the plone_portal_state multi adapter.
  • And the third one is the (somehow deprecated?) CMF style to get the portal object.

Here is the code as external method, just copy it in an appropriate place, and add the external method where you want in your Plone site.

from zope.app.component.hooks import getSite
from zope.component import getMultiAdapter
from Products.CMFCore.utils import getToolByName
from time import time as time_now
from StringIO import StringIO

COUNT = 1000

def testGetPlone(self):
    """Small benchmark"""

    out = StringIO()
    request = self.REQUEST
    t0 = time_now()

    for x in xrange(COUNT):
        plone = getSite()
    t1 = time_now()

    for x in xrange(COUNT):
        plone = getMultiAdapter(
            (self, request),
    t2 = time_now()

    for x in xrange(COUNT):
        plone = getToolByName(self, 'portal_url').getPortalObject()
    t3 = time_now()

    print >> out, COUNT, "times getSite Zope 3 function:", t1 - t0, "seconds"
    print >> out, COUNT, "times plone_portal_state multi adapter:", t2 - t1, "seconds"
    print >> out, COUNT, "times old style getToolByName:", t3 - t2, "seconds"
    return out.getvalue()

Yeah and here are the results on my MacBook:

1000 times getSite Zope 3 function: 0.000494956970215 seconds
1000 times plone_portal_state multi adapter: 0.0551428794861 seconds
1000 times old style getToolByName: 0.0382490158081 seconds

What conclusions can we expose here?

  1. The old style and deprecated getToolByName is faster than the new style way that uses the plone_portal_state multi adapter that is supposed to cache the result! I ran the test several times to confirm this is not a bad trip or beer abuse effect. No, this is true, just continue using the old CMF style way to get a tool or the portal.
  2. The (poorly used in Plone) Zope 3 "getSite" does not need a context or request object to get the site object. Wow! Interresting in lots of design situations. And in addition, it is 100 times faster than the "official" APIs.
  3. Zope 3 offers new services that deserve to be used more and more. Plone suffers of serious performances issues, more particularly for authenticated users, sorry to say this but it's true. We shouldn't need to use ZEO or Squid when 10 authenticated users manage content in a Plone site.

Okay, that benchmark does not match real applications. Who searches 1000 times the portal object in a Plone based app?

But there are certainly dozen of other tricky things in Plone, Archetypes or CMF that should be re designed to address the Plone performance issues.

Have you guys found other hints to have a faster Plone without the need to cache somewhere?

Comments !