Merge pull request #31 from rafaelkallis/copilot/add-benchmark-space-mem-used

Add memory benchmarking with valgrind massif

Author: rafaelkallis

CMakeLists.txt

@@ -74,3 +74,9 @@ add_executable(bench
   "${PROJECT_SOURCE_DIR}/bench/query_sparse_zipf.cpp"
   )
 target_link_libraries(bench art picobench zipf)
+
+# bench-mem executable (for memory profiling with valgrind)
+add_executable(bench-mem
+  "${PROJECT_SOURCE_DIR}/bench/memory.cpp"
+  )
+target_link_libraries(bench-mem art zipf)

Makefile

@@ -16,7 +16,25 @@ test:
 bench:
 	@if [ -f ./$(BUILD_DIR)/bench ]; then ./$(BUILD_DIR)/bench; else echo "Please run 'make' or 'make release' first" && exit 1; fi
 
+bench-mem:
+	@if [ ! -f ./$(BUILD_DIR)/bench-mem ]; then echo "Please run 'make' or 'make release' first" && exit 1; fi
+	@echo "Running memory benchmark with uniform distribution..."
+	@valgrind --tool=massif --massif-out-file=./$(BUILD_DIR)/massif.out.uniform ./$(BUILD_DIR)/bench-mem uniform
+	@ms_print ./$(BUILD_DIR)/massif.out.uniform | head -50
+	@echo ""
+	@echo "Running memory benchmark with Zipfian distribution..."
+	@valgrind --tool=massif --massif-out-file=./$(BUILD_DIR)/massif.out.zipf ./$(BUILD_DIR)/bench-mem zipf
+	@ms_print ./$(BUILD_DIR)/massif.out.zipf | head -50
+	@echo ""
+	@echo "Full reports saved to:"
+	@echo "  - ./$(BUILD_DIR)/massif.out.uniform"
+	@echo "  - ./$(BUILD_DIR)/massif.out.zipf"
+	@echo ""
+	@echo "To view full reports, run:"
+	@echo "  ms_print ./$(BUILD_DIR)/massif.out.uniform"
+	@echo "  ms_print ./$(BUILD_DIR)/massif.out.zipf"
+
 clean:
 	rm -rf ./$(BUILD_DIR)
 
-.PHONY: test bench
+.PHONY: test bench bench-mem

README.md

@@ -44,6 +44,9 @@ make test
 
 # run benchmarks
 make bench
+
+# run memory benchmarks (requires valgrind)
+make bench-mem
 ```
 
 ## Benchmark results 
@@ -101,6 +104,31 @@ query sparse zipf:
 
 You can replicate using `make release && make bench`
 
+## Memory Benchmark
+
+The `make bench-mem` command runs memory profiling using [Valgrind Massif](https://valgrind.org/docs/manual/ms-manual.html) to measure the memory footprint of the ART data structure. The benchmark tests two scenarios:
+
+1. **Uniform distribution**: Keys are uniformly distributed
+2. **Zipfian distribution**: Keys follow a Zipfian distribution (realistic workload with hot keys)
+
+Both benchmarks insert 1,000,000 elements with `nullptr` values to measure only the data structure overhead.
+
+Example usage:
+```bash
+# Build and run memory benchmarks
+make release && make bench-mem
+```
+
+The command will output a summary of memory usage for both distributions and save detailed reports to:
+- `build/massif.out.uniform` - Full report for uniform distribution
+- `build/massif.out.zipf` - Full report for Zipfian distribution
+
+To view the full reports:
+```bash
+ms_print build/massif.out.uniform
+ms_print build/massif.out.zipf
+```
+
 
 ## References
 

bench/memory.cpp

@@ -0,0 +1,72 @@
+/**
+ * @file memory usage benchmarks
+ * @author Rafael Kallis <rk@rafaelkallis.com>
+ */
+
+#include "art.hpp"
+#include "zipf.hpp"
+#include <cstdint>
+#include <functional>
+#include <iostream>
+#include <string>
+
+using std::string;
+using std::to_string;
+using std::hash;
+
+// Number of elements to insert for memory benchmarks
+const uint32_t NUM_ELEMENTS = 1000000;
+
+/**
+ * Memory benchmark with uniformly distributed keys
+ * Uses nullptr values to measure only data structure overhead
+ */
+static void memory_uniform() {
+  art::art<int*> m;
+  hash<uint32_t> h;
+  
+  // Fill tree with nullptr values
+  for (uint32_t i = 0; i < NUM_ELEMENTS; i++) {
+    m.set(to_string(h(i)).c_str(), nullptr);
+  }
+  
+  std::cout << "Inserted " << NUM_ELEMENTS << " elements (uniform distribution)" << std::endl;
+}
+
+/**
+ * Memory benchmark with Zipfian distributed keys
+ * Uses nullptr values to measure only data structure overhead
+ */
+static void memory_zipf() {
+  art::art<int*> m;
+  hash<uint32_t> h;
+  fast_zipf rng(NUM_ELEMENTS);
+  
+  // Fill tree with nullptr values
+  for (uint32_t i = 0; i < NUM_ELEMENTS; i++) {
+    m.set(to_string(h(rng())).c_str(), nullptr);
+  }
+  
+  std::cout << "Inserted " << NUM_ELEMENTS << " elements (zipfian distribution)" << std::endl;
+}
+
+int main(int argc, char *argv[]) {
+  if (argc < 2) {
+    std::cerr << "Usage: " << argv[0] << " <uniform|zipf>" << std::endl;
+    return 1;
+  }
+  
+  std::string mode(argv[1]);
+  
+  if (mode == "uniform") {
+    memory_uniform();
+  } else if (mode == "zipf") {
+    memory_zipf();
+  } else {
+    std::cerr << "Unknown mode: " << mode << std::endl;
+    std::cerr << "Valid modes: uniform, zipf" << std::endl;
+    return 1;
+  }
+  
+  return 0;
+}