libcxxdes

libcxxdes is a C++20 Discrete Event Simulation library inspired by SimPy. Just like Simpy, libcxxdes models discrete-event systems using processes and control-flow expressions. While its API feels familiar to SimPy users, being written in C++, it offers up to a few orders of magnitude performance improvement. Furthermore, discrete-event systems modeled using libcxxdes can be later integrated into other DES simulators whose kernels are written in C or C++, such as SystemC, gem5 or OMNET++.

What sets libcxxdes apart from these simulators is its use of modern C++ features to facilitate system modeling, most prominently the newly introduced coroutine support.

A Quick Example (producer_consumer.cpp)

The following example demonstrates the modeling of an M/M/1 queueing system:

using namespace cxxdes::core;
using namespace cxxdes::core::time_ops;

CXXDES_SIMULATION(producer_consumer_example) {
    /* ... */
    
    coroutine<> producer() {
        for (std::size_t i = 0; i < n_packets; ++i) {
            // places an item to the queue
            co_await q.put(now_seconds());

            // models the arrival time (exponential random variable)
            
            // env.timeout ensures that lambda() is in time units of the
            // environment. Equivalent to lambda() * 1_s in this case.
            co_await env.timeout(lambda());
        }
    }

    coroutine<> consumer() {
        std::size_t n = 0;

        while (true) {
            // blocks until an item is in the queue
            auto x = co_await q.pop();
            ++n;

            if (n == n_packets) {
                // end of the experiment
                // calculate the average latency
                avg_latency = total_latency / n_packets;
                co_return ;
            }
            
            // models the service time (exponential random variable)
            co_await env.timeout(mu());

            total_latency += (now_seconds() - x);
        }
    }

    coroutine<> co_main() {
        // start both producer() and consumer() in parallel
        co_await (producer() && consumer());
    }
};

Features

libcxxdes tries to provide the complete feature set of SimPy, it currently supports/provides:

1. Processes modeled using coroutine<T>s.
2. Control flow expressions:
   • Parallel compositions: co_await all_of(p1(), p2(), ...), co_await (p1() && p2), co_await any_of(p1(), p2(), ...), co_await (p1() || p2())
   • Sequential compositions: co_await sequential(p1(), p2(), ...), co_await (p1(), p2(), ...)
   • Timeouts: co_await delay(5), co_await timeout(5_s)
3. Interruptable coroutines which are useful for modelling preemptive resources. --Not feasible to implement. Use || with a queue, same thing.
4. Priority-scheduling of events that take place at the same simulation time. coroutine<T> can be assigned priorities! (lower the number, higher the priority)
5. time_unit() and time_precision() functions for mapping simulation time (integer) to real-world time.
6. Synchronization primitives, such as mutex, semaphore, queue<T>, and event.
7. RAII-style acquisition of resources using co_with (resource) { /* */ } syntax.
8. SimPy-compatible resource, container and preemptive_resource.
9. Debugging facilities, such as getting the stack traces of coroutine<T>s.
10. A template-metaprogramming-based DSL to describe time accurately without suffering from the quirks of the floating-point aritmetic. 1_s + 500_ms + 100_us is mapped to 1'500'100 simulation if the precision is set to 1_us or to 1'500 for a precision of 1_ms.
11. A CMake-based build system to facilitate integrating withn other projects.