Rain is a statistics-based workload generation toolkit that uses parameterized and empirical distributions to model the different classes of workload variations.

• Highly flexible and customizable workload characterization
  • Supports variations in the amount of load, the mix of operations and variations in data popularity (data hotspots)
  • Application-specific request Generators can easily use parameterized or empirical probability distributions to characterize different variations in load
  • Benchmarks configured by simple JSON configuration files
• Modular design of the harness
  • Simple Generator API allowing easy integration of application-specific request Generators that target new systems/applications. Example request generator for the Olio Web 2.0 application used in the Cloudstone Benchmark described and evaluated in our tech report (see Related Documents below for more details)
• Supports closed, open, and partly open loop workload generation
  • Use of the Command pattern (GoF) to implement parameterizable Operations. This design choice allows any thread to execute an Operation opening up the possibility of open loop or partly open-loop workload generation
  • Two thread pools, one containing dedicated load threads based on the maximum number of clients/users to emulate during an experiment and a (possibly smaller) shared thread pool for simulating asynchronous Operation-execution
• Separation of request generation from request execution
  • This design choice allows Rain to produce traces that can be consumed by a suitable third-party high performance load-delivery client e.g., httperf, if desired or feasible
• Multi-track load generation
  • Generate a workload that targets multiple applications/tenants in a shared cluster in a single experiment. This feature is specifically to enable evaluating resource allocation strategies for multi-application/multi-tenant scenarios

• A “workload server” with Ant.

1. SSH into the “workload server”.
2. Checkout Rain: git clone git://github.com/yungsters/rain-workload-toolkit.git
3. Change into the Rain directory: cd rain-workload-toolkit
4. Compile the source code into rain.jar: ant package
5. Bundle the workload you want to use into a JAR.
   • To compile the Olio package, use ant package-olio to create workloads/olio.jar.
   • You will have to modify the ant build file for any custom workloads.
6. Modify configuration files to describe a run (see Anatomy of a Configuration File).
7. Run the benchmark using: java -Xmx1g -Xms256m -cp "rain.jar:<path to workload JAR>" radlab.rain.Benchmark <path to Rain JSON configuration file>