-
Notifications
You must be signed in to change notification settings - Fork 12
Benchmarking
abhi11 edited this page Nov 3, 2014
·
15 revisions
###Stateless Topology(without Redis)###
###Stateful Topology(with Redis)###
###Explanation###
- Both graphs plot number of tuples processed on the Y-axis and time on the X-axis
- Thus the graphs show rate of processing of tuples
- We can see that the rate of processing for first graph(stateless topology) is lesser than the second graph(stateful topology)
- Thus we can say the stateful strategy using redis is more efficient where following cases apply:
- No data loss(i.e guaranteed message processing)
- Probability of failure is high
###Workflow###
- We have used codahale metrics for benchmarking
- First we setup carbon and graphite
- Then we implement the metric for our bolt. See here
- There are two different Meters, one for processed tuples and one for failed tuples
- We fail some tuples on purpose, so that they are replayed back
- Setting up graphite-carbon and sending metrics from storm can be looked up from here