Adaptive-solutions experimentation runtime (ASER)

ASER is a runtime to do research on adaptive solutions for optimizing the usage of shared resources in a computing system. The runtime is capable of managing the execution of multiple processes, while monitoring the execution and taking decisions to optimize a certain metric of interest.

Execution managers and execution monitors are the two main constructs in ASER. Execution managers are responsible of managing all the life-cycle of the processes belonging to the workload to be monitored---from the creation of these processes to detecting when they have finished. Execution monitors are used to perform a diverse set of tasks that monitor and control the processes being executed. Users can extend the base execution monitor to implement their custom optimization policies.

Minimalistic example

pt::ptree properties;
pt::read_json("config.json", properties);

std::vector<simple_manager::benchmark> benchs = {
  {0, {"/usr/bin/env", "sleep", "5"}},
  {1, {"/usr/bin/env", "sleep", "2"}}
};

simple_manager exec_mgr(properties, benchs);
exec_mgr.start();

This code snippet uses an execution manager (simple_manager) to execute two processes that will run the command sleep---with different arguments. Once the first process finishes, the execution manager will kill the remaining process and the execution will end.

The contents of the configuration file (config.json) are:

{
  "exec_monitor": {
    "type": "simple",
    "sampling_length": 500
  }
}

ASER will create an execution monitor belonging to type "simple", with a sampling period of 500 ms. This simplistic monitor does nothing but run a loop, sleeping for 500 ms on each iteration.

Build instructions

ASER requires a relatively recent compiler with C++14 support. CXX flag can be used to specify the actual C++ compiler to use. The following libraries are also required:

• Boost
• Google Test (for testing purposes only)

The building process uses SCons. If the required libraries are installed in default locations, just running SCons on the root directory will build ASER. Otherwise, BOOST_PATH flag should be used to specify the path were Boost is installed. The resulting binary will be placed in a directory named build. In order to build and run the test cases, use scons test instead.

Following the advice in the documentation, Google Test is downloaded into a directory named 'googletest'. Doing so prevents issues when the compiler used to compile Google Test and ASER are not the same.