What We Learned About Scaling with Apache Storm

It seems like an odd decision to invest in building your own framework to get around a 20% performance improvement on your existing system (80k -> 100k+). Was there really no way to get better performance by tuning the existing system, or throwing more hardware at the problem?

I'd love to read about what else was tried before the decision to abandon Storm was made.

This is sorely lacking in details. Is that really all they did to try and tune storm? As in - nothing? I would be interested to see what other direction they could have gone in with Storm to scale it better for their workloads.

Additionally, it doesn't say much of anything about "what we learned about storm". It just says they stopped using it. There isn't really a good, generalized takeaway here about the system as a whole. This deserves more in-depth analysis.