diff --git a/bagua-core/bagua-core-internal/src/comm_ops/decentralized_full_precision_asynchronous.rs b/bagua-core/bagua-core-internal/src/comm_ops/decentralized_full_precision_asynchronous.rs
index c6b1abdc4..6250f004d 100644
--- a/bagua-core/bagua-core-internal/src/comm_ops/decentralized_full_precision_asynchronous.rs
+++ b/bagua-core/bagua-core-internal/src/comm_ops/decentralized_full_precision_asynchronous.rs
@@ -77,7 +77,6 @@ impl CommOpTrait for DecentralizedFullPrecisionAsynchronous {
              temp_tensor.clone_from(&t.raw, torch_stream as u64);
 
              let src_ready_event = CUDA_EVENT_POOL.take().event;
-             let dst_ready_event = CUDA_EVENT_POOL.take().event;
 
              unsafe {
                  cpp::cpp!([
@@ -100,10 +99,18 @@ impl CommOpTrait for DecentralizedFullPrecisionAsynchronous {
                 PeerSelectionMode::ShiftOne => {
                     unimplemented!()
                 }
-            };
+             };
+
+             let comm_ready_event = CUDA_EVENT_POOL.take().event;
 
              unsafe {
-                  cpp::cpp!([comm_stream as "cudaStream_t"] { CUDACHECK(cudaStreamSynchronize(comm_stream)); });
+                 cpp::cpp!([
+                     comm_ready_event as "cudaEvent_t",
+                     comm_stream as "cudaStream_t"]
+                 {
+                     CUDACHECK(cudaEventRecord(comm_ready_event, comm_stream));
+                     CUDACHECK(cudaEventSynchronize(comm_ready_event));
+                 });
              }
 
              if c.check_abort() {
@@ -111,6 +118,7 @@ impl CommOpTrait for DecentralizedFullPrecisionAsynchronous {
                  return
              }
 
+
              // do we need to wait default stream?
              unsafe {
                  cpp::cpp!([
diff --git a/bagua-core/bagua-core-internal/src/communicators/mod.rs b/bagua-core/bagua-core-internal/src/communicators/mod.rs
index 47f711add..e1d6bc4b7 100644
--- a/bagua-core/bagua-core-internal/src/communicators/mod.rs
+++ b/bagua-core/bagua-core-internal/src/communicators/mod.rs
@@ -3,7 +3,7 @@ use crate::datatypes::{
 };
 use crate::BaguaCoreError;
 use itertools::Itertools;
-use parking_lot::Mutex;
+use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::Arc;
 
 #[derive(Clone, Debug)]
@@ -13,7 +13,7 @@ pub struct BaguaCommunicatorInner {
     pub rank: usize,
     pub nranks: usize,
     pub device_id: usize,
-    pub aborted: Arc<Mutex<bool>>,
+    pub aborted: Arc<AtomicBool>,
 }
 
 #[derive(Clone, Debug)]
@@ -54,7 +54,7 @@ impl BaguaSingleCommunicator {
                 rank,
                 nranks,
                 device_id,
-                aborted: Arc::new(Mutex::new(false)),
+                aborted: Arc::new(AtomicBool::new(false)),
             }),
         }
     }
@@ -456,7 +456,8 @@ impl BaguaCommunicatorInner {
     pub fn abort(&self) {
         let communicator_ptr = self.comm_ptr;
 
-        *self.aborted.lock() = true;
+        self.aborted.store(true, Ordering::Relaxed);
+
         unsafe {
             cpp::cpp!([communicator_ptr as "Al::NCCLCommunicator*"]
             {
@@ -466,7 +467,7 @@ impl BaguaCommunicatorInner {
     }
 
     pub fn check_abort(&self) -> bool {
-        *self.aborted.lock()
+        self.aborted.load(Ordering::Relaxed)
     }
 
     pub fn broadcast(&self, tensor: &mut dyn RawBaguaTensor, root_rank: i32) {
diff --git a/bagua-core/bagua-core-internal/third_party/Aluminum b/bagua-core/bagua-core-internal/third_party/Aluminum
index edb7e1529..dc4ff784d 160000
--- a/bagua-core/bagua-core-internal/third_party/Aluminum
+++ b/bagua-core/bagua-core-internal/third_party/Aluminum
@@ -1 +1 @@
-Subproject commit edb7e1529498aa4858b362fc052c78afdcd85a68
+Subproject commit dc4ff784dc4201ee1358b1bace2d6b26d70ef9c0