Add max_read_window_size config to prefetcher benchmarks

benchmark/benchmarks/benchmark_config_parser.py

@@ -72,6 +72,7 @@ def get_prefetch_config(self) -> Dict[str, Any]:
         prefetch_cfg = self.cfg.benchmarks.prefetch
         return {
             'max_memory_target': getattr(prefetch_cfg, 'max_memory_target', None),
+            'max_read_window_size': getattr(prefetch_cfg, 'max_read_window_size', None),
         }
 
     def get_crt_config(self) -> Dict[str, Any]:
@@ -83,5 +84,5 @@ def get_crt_config(self) -> Dict[str, Any]:
     def get_client_config(self) -> Dict[str, Any]:
         client_cfg = self.cfg.benchmarks.client
         return {
-            'read_window_size': getattr(client_cfg, 'read_window_size', 2147483648),  # Reaslitic default value 8M/2G?
+            'max_read_window_size': getattr(client_cfg, 'max_read_window_size', 2147483648),  # Realistic default value of 2GiB
         }

benchmark/benchmarks/client_benchmark.py

@@ -34,7 +34,7 @@ def run_benchmark(self) -> Dict[str, Any]:
 
         if self.backpressure:
             subprocess_args.append("--enable-backpressure")
-            if (initial_window_size := self.client_config['read_window_size']) is not None:
+            if (initial_window_size := self.client_config['max_read_window_size']) is not None:
                 subprocess_args.extend(["--initial-window-size", str(initial_window_size)])
 
         if (run_time := self.common_config['run_time']) is not None:

benchmark/benchmarks/prefetch_benchmark.py

@@ -72,6 +72,9 @@ def run_benchmark(self) -> None:
         if not self.common_config['download_checksums']:
             prefetch_env["EXPERIMENTAL_MOUNTPOINT_NO_DOWNLOAD_INTEGRITY_VALIDATION"] = "ON"
 
+        if self.prefetch_config['max_read_window_size'] is not None:
+            prefetch_env["UNSTABLE_MOUNTPOINT_MAX_PREFETCH_WINDOW_SIZE"] = str(self.prefetch_config['max_read_window_size'])
+
         subprocess_env = os.environ.copy() | prefetch_env
         log.debug("Subprocess env: %s", subprocess_env)
 

benchmark/conf/config.yaml

@@ -54,6 +54,7 @@ benchmarks:
 
   prefetch:
     max_memory_target: !!null
+    max_read_window_size: !!null
 
   crt:
     crt_benchmarks_path: !!null
@@ -62,7 +63,7 @@ benchmarks:
     # None 
 
   client_backpressure:
-    read_window_size: !!null #2147483648 
+    max_read_window_size: !!null #2147483648
 
 
 hydra:

mountpoint-s3-client/examples/client_benchmark.rs

@@ -218,7 +218,7 @@ struct CliArgs {
     #[arg(
         long,
         help = "Initial read window size in bytes, used to dictate how far ahead we request data from S3",
-        default_value = "0"
+        default_value = "2147483648"
     )]
     initial_window_size: Option<usize>,
     #[arg(long, help = "Output file to write the results to", value_name = "OUTPUT_FILE")]

mountpoint-s3-fs/examples/prefetch_benchmark.rs

@@ -92,7 +92,7 @@ pub struct CliArgs {
     #[arg(
         long,
         help = "Size of read requests requests to the prefetcher",
-        default_value_t = 128 * 1024,
+        default_value_t = 256 * 1024,
         value_name = "BYTES",
     )]
     read_size: usize,
@@ -233,9 +233,9 @@ fn main() -> anyhow::Result<()> {
         let received_size = received_bytes.load(Ordering::SeqCst);
         total_bytes += received_size;
         println!(
-            "{iteration}: received {received_size} bytes in {:.2}s: {:.2} Gib/s",
+            "{iteration}: received {received_size} bytes in {:.2}s: {:.2} Gb/s",
             elapsed.as_secs_f64(),
-            (received_size as f64) / elapsed.as_secs_f64() / (1024 * 1024 * 1024 / 8) as f64
+            (received_size as f64) / elapsed.as_secs_f64() / (1000 * 1000 * 1000 / 8) as f64
         );
         iter_results.push(json!({
             "iteration": iteration,
@@ -246,9 +246,9 @@ fn main() -> anyhow::Result<()> {
     }
     let total_elapsed = total_start.elapsed();
     println!(
-        "\nTotal: {iteration} iterations, {total_bytes} bytes in {:.2}s: {:.2} Gib/s",
+        "\nTotal: {iteration} iterations, {total_bytes} bytes in {:.2}s: {:.2} Gb/s",
         total_elapsed.as_secs_f64(),
-        (total_bytes as f64) / total_elapsed.as_secs_f64() / (1024 * 1024 * 1024 / 8) as f64
+        (total_bytes as f64) / total_elapsed.as_secs_f64() / (1000 * 1000 * 1000 / 8) as f64
     );
 
     if let Some(output_path) = args.output_file {