rt(alt): tweak some constants to improve scalability (#5935)

This patch aims to reduce the number of threads that get no-op wakeups.

tokio/src/runtime/io/scheduled_io.rs

@@ -17,6 +17,93 @@ use std::task::{Context, Poll, Waker};
 
 /// Stored in the I/O driver resource slab.
 #[derive(Debug)]
+// # This struct should be cache padded to avoid false sharing. The cache padding rules are copied
+// from crossbeam-utils/src/cache_padded.rs
+//
+// Starting from Intel's Sandy Bridge, spatial prefetcher is now pulling pairs of 64-byte cache
+// lines at a time, so we have to align to 128 bytes rather than 64.
+//
+// Sources:
+// - https://www.intel.com/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-optimization-manual.pdf
+// - https://github.com/facebook/folly/blob/1b5288e6eea6df074758f877c849b6e73bbb9fbb/folly/lang/Align.h#L107
+//
+// ARM's big.LITTLE architecture has asymmetric cores and "big" cores have 128-byte cache line size.
+//
+// Sources:
+// - https://www.mono-project.com/news/2016/09/12/arm64-icache/
+//
+// powerpc64 has 128-byte cache line size.
+//
+// Sources:
+// - https://github.com/golang/go/blob/3dd58676054223962cd915bb0934d1f9f489d4d2/src/internal/cpu/cpu_ppc64x.go#L9
+#[cfg_attr(
+    any(
+        target_arch = "x86_64",
+        target_arch = "aarch64",
+        target_arch = "powerpc64",
+    ),
+    repr(align(128))
+)]
+// arm, mips, mips64, riscv64, sparc, and hexagon have 32-byte cache line size.
+//
+// Sources:
+// - https://github.com/golang/go/blob/3dd58676054223962cd915bb0934d1f9f489d4d2/src/internal/cpu/cpu_arm.go#L7
+// - https://github.com/golang/go/blob/3dd58676054223962cd915bb0934d1f9f489d4d2/src/internal/cpu/cpu_mips.go#L7
+// - https://github.com/golang/go/blob/3dd58676054223962cd915bb0934d1f9f489d4d2/src/internal/cpu/cpu_mipsle.go#L7
+// - https://github.com/golang/go/blob/3dd58676054223962cd915bb0934d1f9f489d4d2/src/internal/cpu/cpu_mips64x.go#L9
+// - https://github.com/golang/go/blob/3dd58676054223962cd915bb0934d1f9f489d4d2/src/internal/cpu/cpu_riscv64.go#L7
+// - https://github.com/torvalds/linux/blob/3516bd729358a2a9b090c1905bd2a3fa926e24c6/arch/sparc/include/asm/cache.h#L17
+// - https://github.com/torvalds/linux/blob/3516bd729358a2a9b090c1905bd2a3fa926e24c6/arch/hexagon/include/asm/cache.h#L12
+//
+// riscv32 is assumed not to exceed the cache line size of riscv64.
+#[cfg_attr(
+    any(
+        target_arch = "arm",
+        target_arch = "mips",
+        target_arch = "mips64",
+        target_arch = "riscv32",
+        target_arch = "riscv64",
+        target_arch = "sparc",
+        target_arch = "hexagon",
+    ),
+    repr(align(32))
+)]
+// m68k has 16-byte cache line size.
+//
+// Sources:
+// - https://github.com/torvalds/linux/blob/3516bd729358a2a9b090c1905bd2a3fa926e24c6/arch/m68k/include/asm/cache.h#L9
+#[cfg_attr(target_arch = "m68k", repr(align(16)))]
+// s390x has 256-byte cache line size.
+//
+// Sources:
+// - https://github.com/golang/go/blob/3dd58676054223962cd915bb0934d1f9f489d4d2/src/internal/cpu/cpu_s390x.go#L7
+// - https://github.com/torvalds/linux/blob/3516bd729358a2a9b090c1905bd2a3fa926e24c6/arch/s390/include/asm/cache.h#L13
+#[cfg_attr(target_arch = "s390x", repr(align(256)))]
+// x86, wasm, and sparc64 have 64-byte cache line size.
+//
+// Sources:
+// - https://github.com/golang/go/blob/dda2991c2ea0c5914714469c4defc2562a907230/src/internal/cpu/cpu_x86.go#L9
+// - https://github.com/golang/go/blob/3dd58676054223962cd915bb0934d1f9f489d4d2/src/internal/cpu/cpu_wasm.go#L7
+// - https://github.com/torvalds/linux/blob/3516bd729358a2a9b090c1905bd2a3fa926e24c6/arch/sparc/include/asm/cache.h#L19
+//
+// All others are assumed to have 64-byte cache line size.
+#[cfg_attr(
+    not(any(
+        target_arch = "x86_64",
+        target_arch = "aarch64",
+        target_arch = "powerpc64",
+        target_arch = "arm",
+        target_arch = "mips",
+        target_arch = "mips64",
+        target_arch = "riscv32",
+        target_arch = "riscv64",
+        target_arch = "sparc",
+        target_arch = "hexagon",
+        target_arch = "m68k",
+        target_arch = "s390x",
+    )),
+    repr(align(64))
+)]
 pub(crate) struct ScheduledIo {
     pub(super) linked_list_pointers: UnsafeCell<linked_list::Pointers<Self>>,
 

tokio/src/runtime/scheduler/multi_thread_alt/worker.rs

@@ -817,7 +817,7 @@ impl Worker {
     /// workers will be trying to steal at the same time.
     fn search_for_work(&mut self, cx: &Context, mut core: Box<Core>) -> NextTaskResult {
         #[cfg(not(loom))]
-        const ROUNDS: usize = 1;
+        const ROUNDS: usize = 4;
 
         #[cfg(loom)]
         const ROUNDS: usize = 1;
@@ -1000,6 +1000,14 @@ impl Worker {
             // Number of tasks we want to try to spread across idle workers
             let num_fanout = cmp::min(defer.len(), cx.shared().idle.num_idle(&synced.idle));
 
+            // Cap the number of threads woken up at one time. This is to limit
+            // the number of no-op wakes and reduce mutext contention.
+            //
+            // This number was picked after some basic benchmarks, but it can
+            // probably be tuned using the mean poll time value (slower task
+            // polls can leverage more woken workers).
+            let num_fanout = cmp::min(2, num_fanout);
+
             if num_fanout > 0 {
                 cx.shared()
                     .push_remote_task_batch_synced(&mut synced, defer.drain(..num_fanout));