Storm Management and Maintenance - Real-time Analytics with Storm and Cassandra - page 193

Scenario 2
Here, we have acted on the realization that Bolt B is overloaded and has increased the par-
allelism, as shown in the following figure:
The preceding figure describes one scenario capturing the distribution of various instances
of the bolts and spouts across different nodes in the cluster. Here, we have acted on the
realization that a bolt is overloaded and we observed the capacity, and by brute force, in-
creased the parallelism of only that bolt.
Now, having done this, we have achieved the required parallelism; let's now have a look at
the network latency, in terms of how many tuples are moving between nodes (internode
communication is a mandatory element in a distributed computing setup):
• 50 percent of the traffic is hopping between spouts on Machine 1 and Machine 2
• 50 percent of the traffic is hopping between Machine 1 and Machine 3
• 100 percent of the traffic is hopping between Machine 2 and Machine 3
Now let's see another illustration with a slight variation in the parallelism.