Because there are plenty of environments that could use some stream processing
(monitoring anyone?), but don't have Java in any form, and including JVM would
be a gigantic overhead, in terms of disk space and memory consumption.
Are you talking about IoT? Presumably these devices don't have the large amounts of data locally that would warrant the complexity of a MapReduce framework. You could just run a simple WebSockets process that would do the aggregation in memory.
No, I am talking about regular servers and monitoring them (or rather,
processing data collected from monitoring). Using up a one third of RAM
I planned to allocate for the whole server, just to have a message forwarder
that can't even do anything useful yet is not something to be desired.