Create performance benchmark for Deferreds

[thijs]

James Knight has previously implemented Deferred in Pyrex. The reason this isn't included in Twisted is that we just have no reliable, accurate way to measure what the performance changes of this are. The same applies to cdefer.c in Deferred implementation in C.

It would useful to have a benchmark which makes it possible to determine the utility of a Pyrex or C implementation of Deferred. Without actually measuring, statements about how fast Pyrex, C or pure Python is are worthless.

Deriving something from the code which showed a 5% - 20% speedup might be a good way to start on such a benchmark. It's important to cover all uses of Deferreds here. It's also okay if some parts show a slowdown, as long as the overall improvement is enough. Of course, it's better if everything is faster. :)

(ticket description derived from exarkun's comment in #2245)

[thijs]

defer benchmark from source:branches/cdefer-2245-4/doc/core/benchmarks (#2245)

[thijs]

timer benchmark utility from source:branches/cdefer-2245-4/doc/core/benchmarks (#2245)

[thijs]

I attached defer.py and timer.py from the latest branch for ticket #2245, that run a twisted.internet.defer benchmark, with outputting:

$ python defer.py Disabled GC
twisted.internet.defer
instantiate () 0.205390930176
instantiateShootCallback () 0.780241012573
instantiateShootErrback () 0.100492954254
instantiateAddCallbacksNoResult (10,) 0.00270199775696
instantiateAddCallbacksNoResult (1000,) 0.263583898544
instantiateAddCallbacksNoResult (10000,) 3.02256608009
instantiateAddCallbacksBeforeResult (10,) 0.00721502304077
instantiateAddCallbacksBeforeResult (1000,) 0.712515115738
instantiateAddCallbacksBeforeResult (10000,) 13.2036309242
instantiateAddCallbacksAfterResult (10,) 0.00922584533691
instantiateAddCallbacksAfterResult (1000,) 0.903044939041
instantiateAddCallbacksAfterResult (10000,) 8.97325801849
pauseUnpause (10,) 0.00899577140808
pauseUnpause (1000,) 0.912666082382
pauseUnpause (10000,) 15.3666989803

I suggest adding defer.py (and updating timer.py in the trunk).

[moshez]

Adding branch field

[thijs]

Plans for a benchmark api are described in #3715

[exarkun]

I think that branch field setting is erroneous.

[exarkun]

(In [26577]) Branching to 'defer-benchmarks-3704'

[thijs]

(In [26594]) Add copyright header refs #3704

[exarkun]

I made some trivial tweaks. I think this is good enough to merge. I'm sure we'll want to continue changing things as we get more benchmarking experience.

[exarkun]

(In [26630]) Merge defer-benchmarks-3704

Author: Peaker, thijs, exarkun Reviewer: exarkun Fixes: #3704

Add a benchmark for Deferred performance.