A benchmarking framework for Mesos.
This is work in progress in its early stages. Expect bugs and missing features.
The Mesosaurus framework creates work loads that can be configured to simulate the behavior of other frameworks on a Mesos cluster.
We intend to come up with configurations that approximate Marathon, Chronos, Hadoop, MPI, Storm, Spark, and many others. By running multiple instances of Mesosaurus concurrently, it should be possible to simulate large production systems comprising of several frameworks competing for resources. It is our goal that Metrics from such runs can then be used for cluster and Mesos core performance analysis.
- Poisson-distributed task arrival times.
- Normal/Gauss-distributed task durations and resource consumptions.
- Tasks that claim resource offers and use up computer resources in the amounts specified.
- A configurable load factor for actual CPU load.
- Automated provisioning of the task executor binary.
- Configurable task failure rate. No manual installation steps on any slave nodes necessary.
- Read in external traces from other systems and create tasks to simulate the recreation of past scenarios.
- Store traces.
- Real-time plotting of interesting metrics.
- Install Mesos on a cluster - or simply on a single machine if you just want to try it out. In any case, is easiest to follow the instructions at Mesosphere to do this.
- Download, install and run the latest version of Mesosaurus:
TODO
-
Check out the latest source code from its git repository:
git clone https://github.com/mesosphere/mesosaurus.git
-
Build the task executor, which is written in C++.
cd mesosaurus/task make
-
Download and install libraries that Mesosaurus depends on and translate the Scala source code to bytecode:
cd mesosaurus/ sbt compile
-
If you have a Mesos master running on your local machine, check there is entry in /etc/hosts:
<real machine ip> <hostname>
then, you can simply execute Mesosaurus with default values for all settings:cd mesosaurus/ sbt run
-
As a more complex example, you can run 10 tasks that take on average 1 second and arrive on average every 2 seconds on a Mesos installation with its master at a specified IP address and port:
sbt "run -tasks 10 -duration 1000 -arrival 2000 -master 127.0.0.1:5050"
- Install ScalaIDE for Eclipse.
- Install sbteclipse.
- To create an eclipse project that you can import into Eclipse
for Mesosaurus development or simply for source code browsing:
This creates a file with the name ".project" in the Mesosaurus directory.
sbt eclipse
- Import the Mesosaurus project using the Import Wizard in Eclipse, indicating Existing Projects into Workspace and selecting your Mesosaurus directory as the base of your existing project.
Execute the Mesosaurus program just to peruse a list of its available command line options:
sbt "run -h"
We try to follow the usual Scala Style Guide, but no TAB character is allowed anywhere in source code.