diff --git a/src/platform/waiter_linux.h b/src/platform/waiter_linux.h
index c152fcf..9503596 100644
--- a/src/platform/waiter_linux.h
+++ b/src/platform/waiter_linux.h
@@ -1,6 +1,13 @@
 #pragma once
 
 #include <pthread.h>
+#include <fcntl.h>
+#include <sys/stat.h>
+#include <semaphore.h>
+#include <errno.h>
+
+#include <atomic>
+#include <tuple>
 
 #include "def.h"
 #include "log.h"
@@ -104,98 +111,191 @@ public:
 
 #pragma pop_macro("IPC_PTHREAD_FUNC_")
 
-class semaphore {
-    mutex     lock_;
-    condition cond_;
-    long      counter_;
+class sem_helper {
+public:
+    using handle_t = sem_t*;
+
+    constexpr static handle_t invalid() noexcept {
+        return SEM_FAILED;
+    }
+
+    static handle_t open(char const* name, long count) {
+        handle_t sem = ::sem_open(name, O_CREAT, 0666, count);
+        if (sem == SEM_FAILED) {
+            ipc::error("fail sem_open[%d]: %s\n", errno, name);
+            return invalid();
+        }
+        return sem;
+    }
+
+#pragma push_macro("IPC_SEMAPHORE_FUNC_")
+#undef  IPC_SEMAPHORE_FUNC_
+#define IPC_SEMAPHORE_FUNC_(CALL, PAR)              \
+    if (::CALL(PAR) != 0) {                         \
+        ipc::error("fail " #CALL "[%d]\n", errno);  \
+        return false;                               \
+    }                                               \
+    return true
+
+    static bool close(handle_t h) {
+        if (h == invalid()) return false;
+        IPC_SEMAPHORE_FUNC_(sem_close, h);
+    }
+
+    static bool destroy(char const* name) {
+        IPC_SEMAPHORE_FUNC_(sem_unlink, name);
+    }
+
+    static bool post(handle_t h) {
+        if (h == invalid()) return false;
+        IPC_SEMAPHORE_FUNC_(sem_post, h);
+    }
+
+    static bool wait(handle_t h) {
+        if (h == invalid()) return false;
+        IPC_SEMAPHORE_FUNC_(sem_wait, h);
+    }
+
+#pragma pop_macro("IPC_SEMAPHORE_FUNC_")
+};
+
+class waiter_helper {
+    mutex lock_;
+
+    std::atomic<unsigned> waiting_ { 0 };
+    long counter_ = 0;
 
 public:
-    bool open(long count = 0) {
-        if (lock_.open() && cond_.open()) {
-            IPC_UNUSED_ auto guard = ipc::detail::unique_lock(lock_);
-            counter_ = count;
-            return true;
+    using handle_t = std::tuple<std::string, sem_helper::handle_t, sem_helper::handle_t>;
+
+    static handle_t invalid() noexcept {
+        return std::make_tuple(std::string{}, sem_helper::invalid(), sem_helper::invalid());
+    }
+
+    handle_t open_h(std::string && name) {
+        auto sem = sem_helper::open(("__WAITER_HELPER_SEM__" + name).c_str(), 0);
+        if (sem == sem_helper::invalid()) {
+            return invalid();
         }
-        return false;
+        auto han = sem_helper::open(("__WAITER_HELPER_HAN__" + name).c_str(), 0);
+        if (han == sem_helper::invalid()) {
+            return invalid();
+        }
+        return std::make_tuple(std::move(name), sem, han);
+    }
+
+    void release_h(handle_t const & h) {
+        sem_helper::close(std::get<2>(h));
+        sem_helper::close(std::get<1>(h));
+    }
+
+    void close_h(handle_t const & h) {
+        auto const & name = std::get<0>(h);
+        sem_helper::destroy(("__WAITER_HELPER_HAN__" + name).c_str());
+        sem_helper::destroy(("__WAITER_HELPER_SEM__" + name).c_str());
+    }
+
+    bool open() {
+        return lock_.open();
     }
 
     void close() {
-        cond_.close();
         lock_.close();
     }
 
     template <typename F>
-    bool wait_if(F&& check) {
-        bool ret = true;
-        IPC_UNUSED_ auto guard = ipc::detail::unique_lock(lock_);
-        while ((counter_ <= 0) &&
-               std::forward<F>(check)() &&
-               (ret = cond_.wait(lock_))) ;
-        -- counter_;
+    bool wait_if(handle_t const & h, F&& pred) {
+        waiting_.fetch_add(1, std::memory_order_release);
+        {
+            IPC_UNUSED_ auto guard = ipc::detail::unique_lock(lock_);
+            if (!std::forward<F>(pred)()) return true;
+            ++ counter_;
+        }
+        bool ret = sem_helper::wait(std::get<1>(h));
+        waiting_.fetch_sub(1, std::memory_order_release);
+        ret = sem_helper::post(std::get<2>(h)) && ret;
         return ret;
     }
 
-    template <typename F>
-    bool post(F&& count) {
+    bool notify(handle_t const & h) {
+        std::atomic_thread_fence(std::memory_order_acq_rel);
+        if (waiting_.load(std::memory_order_relaxed) == 0) {
+            return true;
+        }
+        bool ret = true;
         IPC_UNUSED_ auto guard = ipc::detail::unique_lock(lock_);
-        auto c = std::forward<F>(count)();
-        if (c <= 0) return false;
-        counter_ += c;
-        return cond_.broadcast();
+        if (counter_ > 0) {
+            ret = sem_helper::post(std::get<1>(h));
+            -- counter_;
+            ret = ret && sem_helper::wait(std::get<2>(h));
+        }
+        return ret;
+    }
+
+    bool broadcast(handle_t const & h) {
+        std::atomic_thread_fence(std::memory_order_acq_rel);
+        if (waiting_.load(std::memory_order_relaxed) == 0) {
+            return true;
+        }
+        bool ret = true;
+        IPC_UNUSED_ auto guard = ipc::detail::unique_lock(lock_);
+        if (counter_ > 0) {
+            for (long i = 0; i < counter_; ++i) {
+                ret = ret && sem_helper::post(std::get<1>(h));
+            }
+            do {
+                -- counter_;
+                ret = ret && sem_helper::wait(std::get<2>(h));
+            } while (counter_ > 0);
+        }
+        return ret;
     }
 };
 
 class waiter {
-    semaphore sem_;
-    std::atomic<long>     counter_ { 0 };
-    std::atomic<unsigned> opened_  { 0 };
+    waiter_helper helper_;
+    std::atomic<unsigned> opened_ { 0 };
 
 public:
-    using handle_t = waiter * ;
+    using handle_t = waiter_helper::handle_t;
 
-    constexpr static handle_t invalid() {
-        return nullptr;
+    static handle_t invalid() noexcept {
+        return waiter_helper::invalid();
     }
 
     handle_t open(char const * name) {
         if (name == nullptr || name[0] == '\0') {
             return invalid();
         }
-        if ((opened_.fetch_add(1, std::memory_order_acq_rel) == 0) && !sem_.open()) {
+        if ((opened_.fetch_add(1, std::memory_order_acq_rel) == 0) && !helper_.open()) {
             return invalid();
         }
-        return this;
+        return helper_.open_h(name);
     }
 
     void close(handle_t h) {
+        if (h == invalid()) return;
+        helper_.release_h(h);
         if (opened_.fetch_sub(1, std::memory_order_release) == 1) {
-            if (h == invalid()) return;
-            sem_.close();
+            helper_.close_h(h);
+            helper_.close();
         }
     }
 
     template <typename F>
     bool wait_if(handle_t h, F&& pred) {
         if (h == invalid()) return false;
-        counter_.fetch_add(1, std::memory_order_release);
-        IPC_UNUSED_ auto guard = unique_ptr(&counter_, [](decltype(counter_)* c) {
-            c->fetch_sub(1, std::memory_order_release);
-        });
-        return sem_.wait_if(std::forward<F>(pred));
+        return helper_.wait_if(h, std::forward<F>(pred));
     }
 
     void notify(handle_t h) {
         if (h == invalid()) return;
-        sem_.post([this] {
-            return (0 < counter_.load(std::memory_order_relaxed)) ? 1 : 0;
-        });
+        helper_.notify(h);
     }
 
     void broadcast(handle_t h) {
         if (h == invalid()) return;
-        sem_.post([this] {
-            return counter_.load(std::memory_order_relaxed);
-        });
+        helper_.broadcast(h);
     }
 };
 
diff --git a/src/platform/waiter_win.h b/src/platform/waiter_win.h
index eb17156..9b1bb24 100644
--- a/src/platform/waiter_win.h
+++ b/src/platform/waiter_win.h
@@ -8,6 +8,7 @@
 #include "rw_lock.h"
 #include "pool_alloc.h"
 #include "log.h"
+#include "shm.h"
 
 #include "platform/to_tchar.h"
 #include "platform/detail.h"
@@ -37,11 +38,20 @@ public:
     }
 
     bool wait() {
-        return ::WaitForSingleObject(h_, INFINITE) == WAIT_OBJECT_0;
+        DWORD ret;
+        if ((ret = ::WaitForSingleObject(h_, INFINITE)) == WAIT_OBJECT_0) {
+            return true;
+        }
+        ipc::error("fail WaitForSingleObject[%lu]: 0x%08X\n", ::GetLastError(), ret);
+        return false;
     }
 
     bool post(long count = 1) {
-        return !!::ReleaseSemaphore(h_, count, NULL);
+        if (::ReleaseSemaphore(h_, count, NULL)) {
+            return true;
+        }
+        ipc::error("fail ReleaseSemaphore[%lu]\n", ::GetLastError());
+        return false;
     }
 };
 
@@ -62,8 +72,13 @@ class condition {
     mutex     lock_;
     semaphore sema_, handshake_;
 
+    ipc::shm::handle waiting_; // std::atomic<unsigned>
     long * counter_ = nullptr;
 
+    auto waiting_cnt() {
+        return static_cast<std::atomic<unsigned>*>(waiting_.get());
+    }
+
 public:
     friend bool operator==(condition const & c1, condition const & c2) {
         return c1.counter_ == c2.counter_;
@@ -74,9 +89,10 @@ public:
     }
 
     bool open(std::string const & name, long * counter) {
-        if (lock_     .open(name + "__COND_MTX__") &&
-            sema_     .open(name + "__COND_SEM__") &&
-            handshake_.open(name + "__COND_HAN__")) {
+        if (lock_     .open    ("__COND_MTX__" + name) &&
+            sema_     .open    ("__COND_SEM__" + name) &&
+            handshake_.open    ("__COND_HAN__" + name) &&
+            waiting_  .acquire(("__COND_WAITING_CNT__" + name).c_str(), sizeof(std::atomic<unsigned>))) {
             counter_ = counter;
             return true;
         }
@@ -84,6 +100,7 @@ public:
     }
 
     void close() {
+        waiting_  .release();
         handshake_.close();
         sema_     .close();
         lock_     .close();
@@ -91,41 +108,57 @@ public:
 
     template <typename Mutex, typename F>
     bool wait_if(Mutex& mtx, F&& pred) {
+        auto cnt = waiting_cnt();
+        if (cnt != nullptr) {
+            cnt->fetch_add(1, std::memory_order_release);
+        }
         {
             IPC_UNUSED_ auto guard = ipc::detail::unique_lock(lock_);
             if (!std::forward<F>(pred)()) return true;
             ++ *counter_;
         }
         mtx.unlock();
-        bool ret_s = sema_.wait();
-        bool ret_h = handshake_.post();
+        bool ret = sema_.wait();
+        if (cnt != nullptr) {
+            cnt->fetch_sub(1, std::memory_order_release);
+        }
+        ret = handshake_.post() && ret;
         mtx.lock();
-        return ret_s && ret_h;
+        return ret;
     }
 
     bool notify() {
-        bool ret_s = true, ret_h = true;
+        std::atomic_thread_fence(std::memory_order_acq_rel);
+        if (waiting_cnt() != nullptr &&
+            waiting_cnt()->load(std::memory_order_relaxed) == 0) {
+            return true;
+        }
+        bool ret = true;
         IPC_UNUSED_ auto guard = ipc::detail::unique_lock(lock_);
         if (*counter_ > 0) {
-            ret_s = sema_.post();
+            ret = sema_.post();
             -- *counter_;
-            ret_h = handshake_.wait();
+            ret = ret && handshake_.wait();
         }
-        return ret_s && ret_h;
+        return ret;
     }
 
     bool broadcast() {
-        bool ret_s = true, ret_h = true;
+        std::atomic_thread_fence(std::memory_order_acq_rel);
+        if (waiting_cnt() != nullptr &&
+            waiting_cnt()->load(std::memory_order_relaxed) == 0) {
+            return true;
+        }
+        bool ret = true;
         IPC_UNUSED_ auto guard = ipc::detail::unique_lock(lock_);
         if (*counter_ > 0) {
-            ret_s = sema_.post(*counter_);
+            ret = sema_.post(*counter_);
             do {
                 -- *counter_;
-                bool rh = handshake_.wait();
-                ret_h = ret_h && rh;
+                ret = ret && handshake_.wait();
             } while (*counter_ > 0);
         }
-        return ret_s && ret_h;
+        return ret;
     }
 };
 
diff --git a/src/platform/waiter_wrapper.h b/src/platform/waiter_wrapper.h
index 1b0a558..8f9b55e 100644
--- a/src/platform/waiter_wrapper.h
+++ b/src/platform/waiter_wrapper.h
@@ -24,7 +24,7 @@ class condition_impl : public ipc::detail::condition {
 
 public:
     bool open(std::string const & name) {
-        if (h_.acquire((name + "__COND_CNT__").c_str(), sizeof(long volatile))) {
+        if (h_.acquire((name + "__COND_CNT__").c_str(), sizeof(long))) {
             return ipc::detail::condition::open(name, static_cast<long *>(h_.get()));
         }
         return false;
@@ -69,8 +69,9 @@ public:
         if (!h_.acquire(name, sizeof(info_t))) {
             return false;
         }
-        if ((static_cast<info_t*>(h_.get())->opened_.fetch_add(1, std::memory_order_acq_rel) == 0) &&
-            !static_cast<info_t*>(h_.get())->object_.open(std::forward<P>(params)...)) {
+        auto info = static_cast<info_t*>(h_.get());
+        if ((info->opened_.fetch_add(1, std::memory_order_acq_rel) == 0) &&
+            !info->object_.open(std::forward<P>(params)...)) {
             return false;
         }
         return true;
@@ -78,8 +79,9 @@ public:
 
     void close() {
         if (!h_.valid()) return;
-        if (static_cast<info_t*>(h_.get())->opened_.fetch_sub(1, std::memory_order_release) == 1) {
-            static_cast<info_t*>(h_.get())->object_.close();
+        auto info = static_cast<info_t*>(h_.get());
+        if (info->opened_.fetch_sub(1, std::memory_order_release) == 1) {
+            info->object_.close();
         }
         h_.release();
     }
@@ -91,17 +93,6 @@ public:
     bool unlock() { return object().unlock(); }
 };
 
-class semaphore_impl : public object_impl<ipc::detail::semaphore> {
-public:
-    bool wait() {
-        return object().wait_if([] { return true; });
-    }
-
-    bool post(long count) {
-        return object().post([count] { return count; });
-    }
-};
-
 class condition_impl : public object_impl<ipc::detail::condition> {
 public:
     bool wait(mutex_impl& mtx) {
@@ -112,6 +103,53 @@ public:
     bool broadcast() { return object().broadcast(); }
 };
 
+class semaphore_impl {
+    sem_helper::handle_t h_;
+    ipc::shm::handle     opened_; // std::atomic<unsigned>
+    std::string          name_;
+
+    auto cnt() {
+        return static_cast<std::atomic<unsigned>*>(opened_.get());
+    }
+
+public:
+    bool open(char const * name, long count) {
+        name_ = name;
+        if (!opened_.acquire(("__SEMAPHORE_IMPL_CNT__" + name_).c_str(), sizeof(std::atomic<unsigned>))) {
+            return false;
+        }
+        if ((h_ = sem_helper::open(("__SEMAPHORE_IMPL_SEM__" + name_).c_str(), count)) == sem_helper::invalid()) {
+            return false;
+        }
+        cnt()->fetch_add(1, std::memory_order_acq_rel);
+        return true;
+    }
+
+    void close() {
+        if (h_ == sem_helper::invalid()) return;
+        sem_helper::close(h_);
+        if (cnt() == nullptr) return;
+        if (cnt()->fetch_sub(1, std::memory_order_release) == 1) {
+            sem_helper::destroy(("__SEMAPHORE_IMPL_SEM__" + name_).c_str());
+        }
+        opened_.release();
+    }
+
+    bool wait() {
+        if (h_ == sem_helper::invalid()) return false;
+        return sem_helper::wait(h_);
+    }
+
+    bool post(long count) {
+        if (h_ == sem_helper::invalid()) return false;
+        bool ret = true;
+        for (long i = 0; i < count; ++i) {
+            ret = ret && sem_helper::post(h_);
+        }
+        return ret;
+    }
+};
+
 } // namespace detail
 } // namespace ipc
 
diff --git a/src/prod_cons.h b/src/prod_cons.h
index 4acd4da..21e6ecf 100644
--- a/src/prod_cons.h
+++ b/src/prod_cons.h
@@ -266,7 +266,7 @@ struct prod_cons_impl<wr<relat::multi , relat::multi, trans::broadcast>> {
         ct_.store(nxt_ct = cur_ct + 1, std::memory_order_release);
         std::forward<F>(f)(&(el->data_));
         // set flag & try update wt
-        el->f_ct_.store(~static_cast<flag_t>(cur_ct));
+        el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
         return true;
     }