VClinic

A portable and efficient infrastracture for value profilers. The document for API and developing a new tool client is described in here.

Pre-built Dockers

One can access to our publicly available docker images for X86 and ARM platforms on zenodo. For detailed usage of the docker images, please refer to the README attached with the image.

Pre-request

• Tested System: Ubuntu 20.04 LTS
• Tested Platform:
  • Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHz
  • ThunderX2 99xx (AArch64)
• Dependencies:
  • CMake >= 3.7
  • GCC/G++

Building VClinic

Following instructions can clone and build VClinic with multiple example client tools:

git clone --recursive /~https://github.com/VClinic/VClinic.git
cd VClinic
./build.sh

Testing

Pre-request

• Python2
• GCC/GFortran

Running benchmarks

We use Nas Parallel Benchmark (NPB 3.4.2) for evaluating the capability as well as the runtime and memory overheads of VClinic. For simplicity, we provide a script to run all the benchmarks with built-in example tools and collect the execution time and peak memory.

# evaluate time and memory overheads for all benchmarks and tools
./run_benchmarks.sh
# evaluate scalabilities
./run_benchmarks_scale.sh

During execution, the script will automatically download the NPB 3.4.2 benchmarks and compile for evaluation. After data collection, the raw data is located in NPB-3.4.2/NPB3.4-OMP/run-<date>/.