SMEC edge-server resource management and scheduling for CPU/GPU intensive workloads.
Includes prober/report receivers (probe/RTT/app), CPU/GPU schedulers (affinity control, CUDA priority control), the entry main.cpp
- Linux,
g++with C++17 support makepthreadand standard build tools
make # builds server_scheduler
make clean # cleans artifactsRun the binary directly:
./server_scheduler [options]Common options:
--probe-port PORT(default 20000)--rtt-port PORT(default 20004)--rtt 0|1(default 0;1enables RTT receiver and disables slack scheduler)--disabledisable CPU scheduling so all apps share all CPUs (mutually exclusive with--rtt 1)--numa-node NODErestrict CPUs to a specific NUMA node--max-cpus COUNTlimit number of CPUs used
Convenience script to build then run:
python3 run.py [--rtt 0|1]