Andreas Jung wrote at 2008-11-23 09:24 +0100:
>> Have you profiled an individual request to learn what the time is spent for?
>> If the time is dominated by client side activity, you need client side
>> skaling to increase the throughput significantly.
>> If, on the other hand, the time is heavily dominated by server side activity,
>> you would need backend skaling (which is currently quite difficult
>> to get) or get the backend significantly faster (I assume that
>> the bandwidth is not the limiting factor).
>This issue is independent of the client-side. ab2 and cmf/plone were
>running on the same (fast) machine.
The other responses you have gotten indicate that client side
activity (especially indexing) can significantly influence
the observed "write performance" (going down by one order of magnitude).
That means: it might be worth to check explicitely that this is not
the case in your setup.
If indexing is suppressed/delayed (as indicated in the other responses),
the number of "store" operations also decrease. Each "store" operation
may account for a few ms (I have observed a general ZEO overhead
of about 3 ms -- however some years ago and therefore on an older
hardware geneation; things may be faster nowadays).
Looking at the number of your "store"s (and the time they take),
you may check whether this may account for the pure write performance
In this case, optimizations would be possible. In principle, the
"store"s could be batched rather than transmitted individually (of course,
this would require ZODB modifications).
Zope-CMF maillist - Zope-CMF@lists.zope.org
See https://bugs.launchpad.net/zope-cmf/ for bug reports and feature requests