[Twisted-Python] Performance issue in reactor.callLater

[Glyph Lefkowitz]

On Wed, 2004-09-08 at 03:58, Stefan Behnel wrote:

> Just to repeat myself: what really get's slower is runUntilCurrent,
> but the gain in callLater should compensate that. So, for the case
> where each callLater leads to a delayed call, the performance gain
> should be marginal (my patch already avoids shifting the list in
> linear time) and in all cases where there are more calls to callLater,
> reset and cancel (added up) than launched delayed calls, I expect the
> heap implementation to be faster by orders of magnitude.

runUntilCurrent happens every tick; callLater only happens sometimes.
Rather than say "average" applications: In my understanding, this will
speed up applications that have lots of timed events scheduled at the
expense of applications that transfer data over the network.

Anthony, if you're listening, do you have a profiling setup that could
give us some authoritative numbers?