Predator-Prey Simulation (PPS)

An actor-based parallel predator-prey ecosystem simulation using MPI. Grid cells are modelled as actors communicating through asynchronous message passing, scheduled by MPI software entities in a 1:N MPI-to-actor relationship.

Design

• Cell-as-actor: Each grid cell is an actor with encapsulated state (predator/prey counts, spatial coordinates)
• 1:N MPI-to-actor: Each MPI rank manages multiple cell actors via CellMPI
• Asynchronous messaging: Cross-rank animal movement uses MPI_Bsend (fire-and-forget)
• Clock actor: Coarse-grained day synchronisation via SUNRISE/SUNSET messages on rank 0
• Out-of-lockstep: Ranks advance sub-steps independently within each day
• Bounce-back: Animals rejected by full cells are returned to their origin via feedback messages

Parameters

Defined in cell/lotka_volterra.h:

#define MAX_PER_CELL 100
#define INITIAL_PREY 10000
#define INITIAL_PREDATORS 5000
#define STEPS_PER_DAY 50
#define DAYS 20

#define ALPHA 0.06     // prey birth rate
#define BETA  0.01     // predation rate
#define DELTA 0.1      // predator reproduction per prey eaten
#define GAMMA 0.04     // predator death rate

Compilation

make            # standard build
make CXXFLAGS="-std=c++17 -O2 -DVERIFY"   # with verification counters
make CXXFLAGS="-std=c++17 -O2 -DDEBUG"    # with debug prints
make clean      # remove build artifacts

Running

# Single rank
srun --ntasks=1 ./main <grid_x> <grid_y>

# Multiple ranks (e.g. 16 ranks on Cirrus)
srun --time=00:10:00 \
     --partition=standard \
     --qos=short \
     --distribution=block:block \
     --hint=nomultithread \
     --account=<budget_code> \
     --nodes=1 \
     --ntasks=16 \
     ./main 200 200

# 512 ranks across 2 nodes
srun --time=00:10:00 \
     --partition=standard \
     --qos=short \
     --exclusive \
     --distribution=block:block \
     --hint=nomultithread \
     --account=<budget_code> \
     --nodes=2 \
     --ntasks=512 \
     ./main 200 200

Scaling Results

Benchmarked on Cirrus (dual AMD EPYC 9825, 288 cores/node).

Strong Scaling (200x200 grid, fixed problem size)

Ranks	Time (s)	Speedup
1	619.51	1.00x
2	532.00	1.16x
4	255.38	2.43x
8	201.41	3.08x
16	118.66	5.22x
32	89.73	6.90x
64	68.33	9.07x
128	51.37	12.06x

Weak Scaling (~2500 cells/rank, proportional animals)

Ranks	Grid	Time (s)
1	50x50	37.55
2	70x70	66.09
4	100x100	110.09
8	140x140	89.07
16	200x200	109.26
32	280x280	145.42
64	400x400	175.75
128	566x566	240.89

Conditional Compilation Flags

Flag	Purpose
-DDEBUG	Enable detailed print statements via DBG_PRINT macro
-DVERIFY	Enable Test class for send/receive count verification

Architecture

predator-prey/
├── main.cpp
├── Makefile
├── README.md
├── logs/
├── framework/
│   ├── actor/
│   │   ├── actor.hpp          # Template-based Actor<DataT, MsgT, RspT> base class
│   │   └── actor_err.hpp      # Actor error codes
│   └── mpi/
│       └── mpi_sw.hpp         # MPI software entity base class
├── cell/
│   ├── cell.hpp               # GridCellActor — cell actor with simulation logic
│   ├── cell.cpp
│   ├── cell_msg.hpp           # CellMsg_s struct and message type enums
│   └── lotka_volterra.h       # Simulation and Lotka-Volterra parameters
├── cell_mpi/
│   ├── cell_mpi.hpp           # CellMPI — MPI rank manager, router, scheduler
│   └── cell_mpi.cpp
├── clock/
│   ├── clock.hpp              # ClockActor — temporal synchronisation
│   └── clock.cpp
└── test/
    └── test.hpp               # Test class for communication verification