Empirical Dataset of 1200 hours of Experiment: Reliability and Performance Measurements of Dynamic Routing Applications

This artifact presents the empirical measurements of an experiment with 1200 hours of runtime. We repeated the experiment 200 runs, each for 6 hours. The data folder provides the empirical data, and the plots folder gives the distribution of reliability and performance measurements in 200 runs. 

For reliability, we simulate component crashes and measure the request loss. To simulate component crashes, we generate a random number for each service and router and stop the docker container if the generated number is below a crash probability (0.5% in this study). 

For performance, we consider the round-trip time of requests going through several services until the requests are processed.

The experiment cases are: 

Four levels of call frequency \(cf\): 10, 25, 50, 100 requests/second (r/s)

Three levels of service: 3, 5, 10
the number of services \(n_{serv}\) in a call sequence

Three architectures: 
Central Entity (CE): one router \(n_{rout} = 1\)
Dynamic Routers (DR): three routers \(n_{rout} = 3\)
Sidecar Architecture (SA): 1 router per service \(n_{rout} = n_{serv}\)