twisted.web2 should be as fast or faster than twisted.web

[itamarst]

It's unclear what needs doing in order to speed things up. One specific complaint is from Zope3, where the WSGI code comes in to play.

For web2 to declare victory and leave, however, there is a specific milestone here: we have to create specific benchmarks that indicate that it is equivalent or superior in performance for basic request handling as web.

[Glyph]

Updated description and summary to reflect a more specific (and hopefully achievable) goal.

[therve]

Is there some benchmarks available ? Something really significant ?

I tried to run some tests using the demo.py file from web2 (only the basic rendering), and build a similar test using t.web. Overall results give web2 twice as slow. I see some reasons:

• the log system is expensive (with LogWrapperResource hook). It's a least responsible for 20% of the work in a request handling
• the other 'clear' problem is the use of zope.interface: the iweb.IResponse call seems to be expensive too.
• the other things is more linked to the new architecture: the channel stuff is much more complex (and extensible of course: we know what we pay for).

So what to do now ?

[Jean-Paul Calderone]

Since we're not going to push twisted.web2 closer to being a real thing, its performance probably isn't a major issue. As we port things to twisted.web, we should be sure not to make performance of twisted.web suffer too much. That's just a general thing, though. I don't know if it needs a ticket (although perhaps we could use some more twisted.web benchmarks so we know if performance is suffering).