Integrate async IO

bpftrace.md

@@ -0,0 +1,63 @@
+# bpftrace: Per-block device read/write byte totals
+
+```bash
+sudo bpftrace -e '
+tracepoint:block:block_rq_complete
+{
+  $bytes = args->nr_sector * 512;
+  /* Cast the string to a char pointer and dereference to get the first byte */
+  $c = *(uint8 *)args->rwbs;
+  $dev = args->dev;
+
+  @read[$dev]  += ($c == 82 ? $bytes : 0);   /* 82 is "R" */
+  @write[$dev] += ($c == 87 ? $bytes : 0);   /* 87 is "W" */
+  @total[$dev] += $bytes;
+  @seen[$dev] = 1;
+}
+'
+```
+
+# bpftrace: Block IO bandwidth tracking at 10ms intervals
+
+```bash
+sudo bpftrace -e '
+tracepoint:block:block_rq_complete
+{
+  $bytes = args->nr_sector * 512;
+  /* Cast the string to a char pointer and dereference to get the first byte */
+  $c = *(uint8 *)args->rwbs;
+
+  /* Accumulate bytes for reads and writes */
+  @read_bytes  += ($c == 82 ? $bytes : 0);
+  @write_bytes += ($c == 87 ? $bytes : 0);
+}
+
+interval:ms:10
+{
+  /* Calculate bandwidth in MB/s using integer arithmetic */
+  /* bytes in 10ms -> bytes_per_second = bytes * 100 */
+  /* MB_per_second = (bytes * 100) / 1048576 */
+  $read_bps = @read_bytes * 100;        /* bytes per second */
+  $write_bps = @write_bytes * 100;
+  $read_mbps = $read_bps / 1048576;     /* MB per second (integer division) */
+  $write_mbps = $write_bps / 1048576;
+
+  /* Time in seconds (elapsed is in nanoseconds) */
+  $time_sec = elapsed / 1000000000;
+  $time_ms_remainder = elapsed % 1000000000;
+  $time_ms = $time_ms_remainder / 1000000;
+
+  printf("%d.%03d: Read: %d MB/s, Write: %d MB/s\n", 
+         $time_sec, $time_ms, $read_mbps, $write_mbps);
+
+  /* Reset counters for next interval */
+  @read_bytes = 0;
+  @write_bytes = 0;
+}
+
+END
+{
+  printf("\nTracing complete.\n");
+}
+'
+```

run.sh

@@ -13,6 +13,7 @@ PROFILE_OUTPUT="perf.data"
 FLAMEGRAPH_OUTPUT="flamegraph.svg"
 SEPARATE_VALUES=false
 ENABLE_MEMORY_PROFILING=false
+USE_ASYNC_IO=false
 PCM_MEMORY="/users/proteet/pcm/build/bin/pcm-memory"
 
 # Function to show usage
@@ -88,6 +89,10 @@ while [[ $# -gt 0 ]]; do
             FLAMEGRAPH_OUTPUT="$2"
             shift 2
             ;;
+        --use-async)
+            USE_ASYNC_IO=true
+            shift
+            ;;
         --help|-h)
             show_usage
             exit 0
@@ -151,6 +156,10 @@ if [ "$SEPARATE_VALUES" = true ]; then
     CMD_ARGS="$CMD_ARGS --separate-values"
 fi
 
+if [ "$USE_ASYNC_IO" = true ]; then
+    CMD_ARGS="$CMD_ARGS --use-async"
+fi
+
 echo "Running sorter with parameters:"
 echo "  File size: $FILE_SIZE"
 echo "  Key size: $KEY_SIZE"

src/CMakeLists.txt

@@ -20,12 +20,18 @@ endif()
 
 find_package(OpenMP REQUIRED)
 
+find_library(LIBURING_LIBRARIES uring)
+find_path(LIBURING_INCLUDE_DIRS liburing.h)
+
 # Add executable
 add_executable(sorter main.cpp)
 
 # target_compile_options(sorter PRIVATE -fopenmp)
 target_link_libraries(sorter PUBLIC OpenMP::OpenMP_CXX)
 
+target_include_directories(sorter PRIVATE ${LIBURING_INCLUDE_DIRS})
+target_link_libraries(sorter PRIVATE ${LIBURING_LIBRARIES})
+
 # Enable additional optimizations
 # target_compile_options(sorter PRIVATE -march=native -mtune=native)
 target_compile_options(sorter PRIVATE -mavx512f -mavx512vl)