diff --git a/lib/pls/include/pls/internal/base/backoff.h b/lib/pls/include/pls/internal/base/backoff.h
index 9e7cbf5..45ca814 100644
--- a/lib/pls/include/pls/internal/base/backoff.h
+++ b/lib/pls/include/pls/internal/base/backoff.h
@@ -14,7 +14,7 @@ namespace internal {
 namespace base {
 
 class backoff {
-  const unsigned long INITIAL_SPIN_ITERS = 2u << 4u;
+  const unsigned long INITIAL_SPIN_ITERS = 2u << 1u;
   const unsigned long MAX_SPIN_ITERS = 2u << 8u;
   const unsigned long MAX_ITERS = 2u << 10u;
   const unsigned long YELD_ITERS = 2u << 10u;
@@ -36,7 +36,7 @@ class backoff {
 
     if (current_ >= YELD_ITERS) {
       PROFILE_LOCK("Yield")
-      this_thread::yield();
+      this_thread::sleep(5);
     }
 
     current_ = std::min(current_ * 2, MAX_ITERS);
diff --git a/lib/pls/include/pls/internal/base/system_details.h b/lib/pls/include/pls/internal/base/system_details.h
index 0f0d5e7..17b51d7 100644
--- a/lib/pls/include/pls/internal/base/system_details.h
+++ b/lib/pls/include/pls/internal/base/system_details.h
@@ -52,15 +52,9 @@ constexpr pointer_t CACHE_LINE_SIZE = 64;
  *
  * Choose the implementation appropriate for your compiler-cpu combination.
  */
-#if (COMPILER == MVCC)
-inline void relax_cpu() {
-  _mm_pause();
-}
-#elif (COMPILER == GCC || COMPILER == LLVM)
 inline void relax_cpu() {
-  asm("pause");
+  asm volatile("pause":: : "memory");
 }
-#endif
 
 }
 }
diff --git a/lib/pls/src/internal/scheduling/abstract_task.cpp b/lib/pls/src/internal/scheduling/abstract_task.cpp
index aeccb1d..85e2c36 100644
--- a/lib/pls/src/internal/scheduling/abstract_task.cpp
+++ b/lib/pls/src/internal/scheduling/abstract_task.cpp
@@ -10,7 +10,7 @@ namespace internal {
 namespace scheduling {
 
 bool abstract_task::steal_work() {
-  thread_local static base::backoff backoff{};
+//  thread_local static base::backoff backoff{};
 
   PROFILE_STEALING("abstract_task::steal_work")
   const auto my_state = base::this_thread::state<thread_state>();
@@ -22,7 +22,7 @@ bool abstract_task::steal_work() {
   for (size_t i = 0; i < max_tries; i++) {
     size_t target = (offset + i) % my_scheduler->num_threads();
     if (target == my_id) {
-      target = (target + 1) % my_scheduler->num_threads();
+      continue;
     }
     auto target_state = my_scheduler->thread_state_for(target);
 
@@ -47,7 +47,7 @@ bool abstract_task::steal_work() {
         if (internal_stealing(current_task)) {
           // internal steal was a success, hand it back to the internal scheduler
           target_state->lock_.reader_unlock();
-          backoff.reset();
+//          backoff.reset();
           return true;
         }
 
@@ -65,7 +65,7 @@ bool abstract_task::steal_work() {
       auto lock = &target_state->lock_;
       if (current_task->split_task(lock)) {
         // top level steal was a success (we did a top level task steal)
-        backoff.reset();
+//        backoff.reset();
         return false;
       }
 
@@ -76,7 +76,8 @@ bool abstract_task::steal_work() {
   }
 
   // internal steal was no success
-  backoff.do_backoff();
+//  backoff.do_backoff();
+//  base::this_thread::sleep(5);
   return false;
 }