Performance Issues while using dyno-queues over dynomite

[krishnaputhran]

I am using Netflix conductor for quite a while. But I am seeing lot of performance issues, especially w.r.t polling. Netflix conductor provides in-memory configuration and dynomite based configuration. When I execute workflows using both the configuration, I see an difference of 4000 ms w.r.t in-memory and dynomite. While worker poll, I see lot polling timeout issues, if the no.of parallel workflows crosses beyond 20. I see people claiming around millions of workflows without any issues, but I am not able to scale beyond 20. So Netflix conductor community redirected here for possible configuration issues. I am using the default configuration provided for Netflix conductor without any changes. So, I like to know whether I need to change any configuration for dynoqueue and dynomite. Any help on this would be appreiated.