Github user revans2 commented on the issue:
https://github.com/apache/storm/pull/2270
@HeartSaVioR
I spent the week end and came up with a prototype that appears to be
working for combining the two, but I want to talk with @Ethanlm and work out
the differences between the two approaches before putting up any kind of patch.
In the current model we are measuring load in terms of how full a queue is
so we can route around the issue using a simple proportional controller
essentially reducing the cost in terms of load for a tuple to be processed, but
I realized we could also measure it in terms the queue sojourn time (how long
each tuple stays in the queue). This patch is also trying to do a cost
reduction controller, with longer term plans were to use ping time for part of
that cost model. If we just go with ping time for the cost model then we now
have everything in terms of time which makes modeling much simpler.
We can also improve it over time as we improve our estimation of how long
it will take for a tuple to go from point A to point B.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
