Merge branch 'master' of https://github.com/Microsoft/multiverso

2016-04-21 13:55:29 +08:00 · 2016-04-21 13:55:29 +08:00 · b856ef8d3a
--- a/Test/main.cpp
+++ b/Test/main.cpp
@ -88,8 +88,6 @@ void TestArray(int argc, char* argv[]) {

  int iter = 1000;

-  if (argc == 2) iter = atoi(argv[1]);
-
  for (int i = 0; i < iter; ++i) {
    // std::vector<float>& vec = shared_array->raw();

--- a/include/multiverso/net/allreduce_engine.h
+++ b/include/multiverso/net/allreduce_engine.h
@ -43,8 +43,8 @@ public:
 */
 enum RecursiveHalvingNodeType {
  Normal, //normal node, 1 group only have 1 machine 
-  ReciveNeighbor, //leader of group when number of machines in this group is 2.
-  SendNeighbor// non-leader machines in group
+  GroupLeader, //leader of group when number of machines in this group is 2.
+  Other// non-leader machines in group
 };

 /*! \brief Network structure for recursive halving algorithm */
--- a/include/multiverso/table_interface.h
+++ b/include/multiverso/table_interface.h
@ -1,6 +1,7 @@
 #ifndef MULTIVERSO_TABLE_INTERFACE_H_
 #define MULTIVERSO_TABLE_INTERFACE_H_

+#include <mutex>
 #include <string>
 #include <unordered_map>
 #include <vector>
@ -40,7 +41,8 @@ public:
 private:
  std::string table_name_;
  int table_id_;
-  std::unordered_map<int, Waiter*> waitings_;
+  std::mutex m_;
+  std::vector<Waiter*> waitings_;
  int msg_id_;
 };

--- a/include/multiverso/util/configure.h
+++ b/include/multiverso/util/configure.h
@ -102,7 +102,12 @@ void SetCMDFlag(const std::string& name, const T& value) {

 #define MV_DECLARE_bool(name)  \
  DECLARE_CONFIGURE(bool, name)
+ 
+#define MV_DEFINE_double(name, default_value, text) \
+  DEFINE_CONFIGURE(double, name, default_value, text)

+#define MV_DECLARE_double(name)  \
+  DECLARE_CONFIGURE(double, name)
 }  // namespace multiverso

 #endif  // MULTIVERSO_UTIL_CONFIGURE_H_
--- a/src/Multiverso.vcxproj.filters
+++ b/src/Multiverso.vcxproj.filters
@ -1,12 +1,6 @@
 <?xml version="1.0" encoding="utf-8"?>
 <Project ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
  <ItemGroup>
-    <ClInclude Include="..\include\multiverso\multiverso.h">
-      <Filter>include</Filter>
-    </ClInclude>
-    <ClInclude Include="..\include\multiverso\table_interface.h">
-      <Filter>include</Filter>
-    </ClInclude>
    <ClInclude Include="..\include\multiverso\table\array_table.h">
      <Filter>table</Filter>
    </ClInclude>
@ -112,11 +106,14 @@
    <ClInclude Include="..\include\multiverso\io\local_stream.h">
      <Filter>io</Filter>
    </ClInclude>
+    <ClInclude Include="..\include\multiverso\multiverso.h">
+      <Filter>system</Filter>
+    </ClInclude>
+    <ClInclude Include="..\include\multiverso\table_interface.h">
+      <Filter>system</Filter>
+    </ClInclude>
  </ItemGroup>
  <ItemGroup>
-    <Filter Include="include">
-      <UniqueIdentifier>{befa27d3-15da-409a-a02e-fc1d9e676f80}</UniqueIdentifier>
-    </Filter>
    <Filter Include="system">
      <UniqueIdentifier>{f1d6488a-2e66-4610-8676-304c070d7b49}</UniqueIdentifier>
    </Filter>
--- a/src/net/allreduce_engine.cpp
+++ b/src/net/allreduce_engine.cpp
@ -123,11 +123,11 @@ void AllreduceEngine::ReduceScatter(char* input, int input_size, int type_size,

  bool is_powerof_2 = (num_machines_ & (num_machines_ - 1)) == 0 ? true : false;
  if (!is_powerof_2) {
-    if (recursive_halving_map_.type == RecursiveHalvingNodeType::SendNeighbor) {
+    if (recursive_halving_map_.type == RecursiveHalvingNodeType::Other) {
      //send local data to neighbor first
      linkers_->Send(recursive_halving_map_.neighbor, input, 0, input_size);
    }
-    else if (recursive_halving_map_.type == RecursiveHalvingNodeType::ReciveNeighbor) {
+    else if (recursive_halving_map_.type == RecursiveHalvingNodeType::GroupLeader) {
      //recieve neighbor data first
      int need_recv_cnt = input_size;
      linkers_->Receive(recursive_halving_map_.neighbor, output, 0, need_recv_cnt);
@ -135,7 +135,7 @@ void AllreduceEngine::ReduceScatter(char* input, int input_size, int type_size,
    }
  }
  //start recursive halfing
-  if (recursive_halving_map_.type != RecursiveHalvingNodeType::SendNeighbor) {
+  if (recursive_halving_map_.type != RecursiveHalvingNodeType::Other) {

    for (int i = 0; i < recursive_halving_map_.k; ++i) {
      int target = recursive_halving_map_.ranks[i];
@ -160,11 +160,11 @@ void AllreduceEngine::ReduceScatter(char* input, int input_size, int type_size,
  int my_reduce_block_idx = rank_;

  if (!is_powerof_2) {
-    if (recursive_halving_map_.type == RecursiveHalvingNodeType::ReciveNeighbor) {
+    if (recursive_halving_map_.type == RecursiveHalvingNodeType::GroupLeader) {
      //send result to neighbor
      linkers_->Send(recursive_halving_map_.neighbor, input, block_start[recursive_halving_map_.neighbor], block_len[recursive_halving_map_.neighbor]);
    }
-    else if (recursive_halving_map_.type == RecursiveHalvingNodeType::SendNeighbor) {
+    else if (recursive_halving_map_.type == RecursiveHalvingNodeType::Other) {
      //receive result from neighbor
      int need_recv_cnt = block_len[my_reduce_block_idx];
      linkers_->Receive(recursive_halving_map_.neighbor, output, 0, need_recv_cnt);
--- a/src/net/allreduce_topo.cpp
+++ b/src/net/allreduce_topo.cpp
@ -5,13 +5,13 @@

 namespace multiverso {

-
 BruckMap::BruckMap() {
  k = 0;
 }

 BruckMap::BruckMap(int n) {
  k = n;
+  // default set to -1
  for (int i = 0; i < n; ++i) {
    in_ranks.push_back(-1);
    out_ranks.push_back(-1);
@ -19,19 +19,21 @@ BruckMap::BruckMap(int n) {
 }

 BruckMap BruckMap::Construct(int rank, int num_machines) {
-  int* dist = new int[num_machines];
+  // distance at k-th communication, distance[k] = 2^k
+  std::vector<int> distance;
  int k = 0;
  for (k = 0; (1 << k) < num_machines; k++) {
-    dist[k] = 1 << k;
+    distance.push_back(1 << k);
  }
  BruckMap bruckMap(k);
  for (int j = 0; j < k; ++j) {
-    int ni = (rank + dist[j]) % num_machines;
-    bruckMap.in_ranks[j] = ni;
-    ni = (rank - dist[j] + num_machines) % num_machines;
-    bruckMap.out_ranks[j] = ni;
+    // set incoming rank at k-th commuication
+    const int in_rank = (rank + distance[j]) % num_machines;
+    bruckMap.in_ranks[j] = in_rank;
+    // set outgoing rank at k-th commuication
+    const int out_rank = (rank - distance[j] + num_machines) % num_machines;
+    bruckMap.out_ranks[j] = out_rank;
  }
-  delete[] dist;
  return bruckMap;
 }

@ -41,9 +43,10 @@ RecursiveHalvingMap::RecursiveHalvingMap() {
 }
 RecursiveHalvingMap::RecursiveHalvingMap(RecursiveHalvingNodeType _type, int n) {
  type = _type;
-  if (type != RecursiveHalvingNodeType::SendNeighbor) {
-    k = n;
+  k = n;
+  if (type != RecursiveHalvingNodeType::Other) {
    for (int i = 0; i < n; ++i) {
+      // defalut set as -1
      ranks.push_back(-1);
      send_block_start.push_back(-1);
      send_block_len.push_back(-1);
@ -52,107 +55,117 @@ RecursiveHalvingMap::RecursiveHalvingMap(RecursiveHalvingNodeType _type, int n)
    }
  }
 }
+
 RecursiveHalvingMap RecursiveHalvingMap::Construct(int rank, int num_machines) {
-  std::vector<RecursiveHalvingMap> rec_maps;
-  for (int i = 0; i < num_machines; ++i) {
-    rec_maps.push_back(RecursiveHalvingMap());
-  }
-  int* distance = new int[num_machines];
-  RecursiveHalvingNodeType* node_type = new RecursiveHalvingNodeType[num_machines];
+  // construct all recursive halving map for all machines
  int k = 0;
-  for (k = 0; (1 << k) < num_machines; k++) {
-    distance[k] = 1 << k;
+  while ((1 << k) <= num_machines) { ++k; }
+  // let 1 << k <= num_machines
+  --k;
+  // distance of each communication
+  std::vector<int> distance;
+  for (int i = 0; i < k; ++i) {
+    distance.push_back(1 << (k - 1 - i));
  }
+
  if ((1 << k) == num_machines) {
+    RecursiveHalvingMap rec_map(RecursiveHalvingNodeType::Normal, k);
+    // if num_machines = 2^k, don't need to group machines
    for (int i = 0; i < k; ++i) {
-      distance[i] = 1 << (k - 1 - i);
-    }
-    for (int i = 0; i < num_machines; ++i) {
-      rec_maps[i] = RecursiveHalvingMap(RecursiveHalvingNodeType::Normal, k);
-      for (int j = 0; j < k; ++j) {
-        int dir = ((i / distance[j]) % 2 == 0) ? 1 : -1;
-        int ni = i + dir * distance[j];
-        rec_maps[i].ranks[j] = ni;
-        int t = i / distance[j];
-        rec_maps[i].recv_block_start[j] = t * distance[j];
-        rec_maps[i].recv_block_len[j] = distance[j];
-      }
+      // communication direction, %2 == 0 is positive
+      const int dir = ((rank / distance[i]) % 2 == 0) ? 1 : -1;
+      // neighbor at k-th communication
+      const int next_node_idx = rank + dir * distance[i];
+      rec_map.ranks[i] = next_node_idx;
+      // receive data block at k-th communication
+      const int recv_block_start = rank / distance[i];
+      rec_map.recv_block_start[i] = recv_block_start * distance[i];
+      rec_map.recv_block_len[i] = distance[i];
+      // send data block at k-th communication
+      const int send_block_start = next_node_idx / distance[i];
+      rec_map.send_block_start[i] = send_block_start * distance[i];
+      rec_map.send_block_len[i] = distance[i];
    }
+    return rec_map;
  }
  else {
-    k--;
+    // if num_machines != 2^k, need to group machines
+
    int lower_power_of_2 = 1 << k;

-    int rest = num_machines - (1 << k);
+    int rest = num_machines - lower_power_of_2;
+
+    std::vector<RecursiveHalvingNodeType> node_type;
    for (int i = 0; i < num_machines; ++i) {
-      node_type[i] = RecursiveHalvingNodeType::Normal;
+      node_type.push_back(RecursiveHalvingNodeType::Normal);
    }
+    // group, two machine in one group, total "rest" groups will have 2 machines.
    for (int i = 0; i < rest; ++i) {
-      int r = num_machines - i * 2 - 1;
-      int l = num_machines - i * 2 - 2;
-      node_type[l] = RecursiveHalvingNodeType::ReciveNeighbor;
-      node_type[r] = RecursiveHalvingNodeType::SendNeighbor;
-    }
-    for (int i = 0; i < k; ++i) {
-      distance[i] = 1 << (k - 1 - i);
+      int right = num_machines - i * 2 - 1;
+      int left = num_machines - i * 2 - 2;
+      // let left machine as group leader
+      node_type[left] = RecursiveHalvingNodeType::GroupLeader;
+      node_type[right] = RecursiveHalvingNodeType::Other;
    }
+    int group_cnt = 0;
+    // cache block information for groups, group with 2 machines will have double block size
+    std::vector<int> group_block_start(lower_power_of_2);
+    std::vector<int> group_block_len(lower_power_of_2, 0);
+    // convert from group to node leader
+    std::vector<int> group_to_node(lower_power_of_2);
+    // convert from node to group
+    std::vector<int> node_to_group(num_machines);

-    int group_idx = 0;
-    int* map_len = new int[lower_power_of_2];
-    int* map_start = new int[lower_power_of_2];
-    int* group_2_node = new int[lower_power_of_2];
-    int* node_to_group = new int[num_machines];
-    for (int i = 0; i < lower_power_of_2; ++i) {
-      map_len[i] = 0;
-    }
    for (int i = 0; i < num_machines; ++i) {
-      if (node_type[i] == RecursiveHalvingNodeType::Normal || node_type[i] == RecursiveHalvingNodeType::ReciveNeighbor) {
-        group_2_node[group_idx++] = i;
-
+      // meet new group
+      if (node_type[i] == RecursiveHalvingNodeType::Normal || node_type[i] == RecursiveHalvingNodeType::GroupLeader) {
+        group_to_node[group_cnt++] = i;
      }
-      map_len[group_idx - 1]++;
-      node_to_group[i] = group_idx - 1;
+      node_to_group[i] = group_cnt - 1;
+      // add block len for this group
+      group_block_len[group_cnt - 1]++;
    }
-    map_start[0] = 0;
+    // calculate the group block start
+    group_block_start[0] = 0;
    for (int i = 1; i < lower_power_of_2; ++i) {
-      map_start[i] = map_start[i - 1] + map_len[i - 1];
+      group_block_start[i] = group_block_start[i - 1] + group_block_len[i - 1];
    }

-    for (int i = 0; i < num_machines; ++i) {
-
-      if (node_type[i] == RecursiveHalvingNodeType::SendNeighbor) {
-        rec_maps[i] = RecursiveHalvingMap(RecursiveHalvingNodeType::SendNeighbor, k);
-        rec_maps[i].neighbor = i - 1;
-        continue;
-      }
-      rec_maps[i] = RecursiveHalvingMap(node_type[i], k);
-      if (node_type[i] == RecursiveHalvingNodeType::ReciveNeighbor) {
-        rec_maps[i].neighbor = i + 1;
-      }
-      for (int j = 0; j < k; ++j) {
-        int dir = ((node_to_group[i] / distance[j]) % 2 == 0) ? 1 : -1;
-        group_idx = group_2_node[(node_to_group[i] + dir * distance[j])];
-        rec_maps[i].ranks[j] = group_idx;
-        int t = node_to_group[i] / distance[j];
-        rec_maps[i].recv_block_start[j] = map_start[t * distance[j]];
-        int tl = 0;
-        for (int tmp_i = 0; tmp_i < distance[j]; ++tmp_i) {
-          tl += map_len[t * distance[j] + tmp_i];
-        }
-        rec_maps[i].recv_block_len[j] = tl;
-      }
+    RecursiveHalvingMap rec_map(node_type[rank], k);
+    if (node_type[rank] == RecursiveHalvingNodeType::Other) {
+      rec_map.neighbor = rank - 1;
+      // not need to construct
+      return rec_map;
    }
-  }
-  for (int i = 0; i < num_machines; ++i) {
-    if (rec_maps[i].type != RecursiveHalvingNodeType::SendNeighbor) {
-      for (int j = 0; j < k; ++j) {
-        int target = rec_maps[i].ranks[j];
-        rec_maps[i].send_block_start[j] = rec_maps[target].recv_block_start[j];
-        rec_maps[i].send_block_len[j] = rec_maps[target].recv_block_len[j];
-      }
+    if (node_type[rank] == RecursiveHalvingNodeType::GroupLeader) {
+      rec_map.neighbor = rank + 1;
    }
+    const int cur_group_idx = node_to_group[rank];
+    for (int i = 0; i < k; ++i) {
+      const int dir = ((cur_group_idx / distance[i]) % 2 == 0) ? 1 : -1;
+      const int next_node_idx = group_to_node[(cur_group_idx + dir * distance[i])];
+      rec_map.ranks[i] = next_node_idx;
+      // get receive block informations
+      const int recv_block_start = cur_group_idx / distance[i];
+      rec_map.recv_block_start[i] = group_block_start[recv_block_start * distance[i]];
+      int recv_block_len = 0;
+      // accumulate block len
+      for (int j = 0; j < distance[i]; ++j) {
+        recv_block_len += group_block_len[recv_block_start * distance[i] + j];
+      }
+      rec_map.recv_block_len[i] = recv_block_len;
+      // get send block informations
+      const int send_block_start = (cur_group_idx + dir * distance[i]) / distance[i];
+      rec_map.send_block_start[i] = group_block_start[send_block_start * distance[i]];
+      int send_block_len = 0;
+      // accumulate block len
+      for (int j = 0; j < distance[i]; ++j) {
+        send_block_len += group_block_len[send_block_start * distance[i] + j];
+      }
+      rec_map.send_block_len[i] = send_block_len;
+    }
+    return rec_map;
  }
-  return rec_maps[rank];
 }

 }
--- a/src/server.cpp
+++ b/src/server.cpp
@ -18,6 +18,7 @@
 namespace multiverso {

 MV_DEFINE_bool(sync, false, "sync or async");
+MV_DEFINE_int(backup_worker_ratio, 0, "ratio% of backup workers, set 20 means 20%");

 Server::Server() : Actor(actor::kServer) {
  RegisterHandler(MsgType::Request_Get, std::bind(
@ -60,25 +61,149 @@ void Server::ProcessAdd(MessagePtr& msg) {
 // If worker k has add delta to server j times when its i-th Get 
 // then the server will return the parameter after all K 
 // workers finished their j-th update
-class SyncServer : public Server {
+
+// TODO(feiga): to delete this, SyncServer is a special case for 
+//              BackupWorkerSyncServer
+
+//class SyncServer : public Server {
+//public:
+//  SyncServer() : Server() {
+//    int num_worker = Zoo::Get()->num_workers();
+//    worker_get_clocks_.reset(new VectorClock(num_worker));
+//    worker_add_clocks_.reset(new VectorClock(num_worker));
+//  }
+//
+//  // make some modification to suit to the sync server
+//  // please not use in other place, may different with the general vector clock
+//  class VectorClock {
+//  public:
+//    explicit VectorClock(int n) : 
+//      local_clock_(n, 0), global_clock_(0), size_(0) {}
+//
+//    // Return true when all clock reach a same number
+//    virtual bool Update(int i) {
+//      ++local_clock_[i];
+//      if (global_clock_ < *(std::min_element(std::begin(local_clock_),
+//        std::end(local_clock_)))) {
+//        ++global_clock_;
+//        if (global_clock_ == *(std::max_element(std::begin(local_clock_),
+//          std::end(local_clock_)))) {
+//          return true;
+//        }
+//      }
+//      return false;
+//    }
+//
+//    std::string DebugString() {
+//      std::string os = "global ";
+//      os += std::to_string(global_clock_) + " local: ";
+//      for (auto i : local_clock_) os += std::to_string(i) + " ";
+//      return os;
+//    }
+//
+//    int local_clock(int i) const { return local_clock_[i]; }
+//    int global_clock() const { return global_clock_; }
+//
+//  protected:
+//    std::vector<int> local_clock_;
+//    int global_clock_;
+//    int size_;
+//  };
+//protected:
+//  void ProcessAdd(MessagePtr& msg) override {
+//    // 1. Before add: cache faster worker
+//    int worker = Zoo::Get()->rank_to_worker_id(msg->src());
+//    if (worker_get_clocks_->local_clock(worker) >
+//        worker_get_clocks_->global_clock()) {
+//      msg_add_cache_.Push(msg);
+//      return;
+//    }
+//    // 2. Process Add
+//    Server::ProcessAdd(msg);
+//    // 3. After add: process cached process get if necessary
+//    if (worker_add_clocks_->Update(worker)) {
+//      CHECK(msg_add_cache_.Empty());
+//      while (!msg_get_cache_.Empty()) {
+//        MessagePtr get_msg;
+//        CHECK(msg_get_cache_.TryPop(get_msg));
+//        int get_worker = Zoo::Get()->rank_to_worker_id(get_msg->src());
+//        Server::ProcessGet(get_msg);
+//        worker_get_clocks_->Update(get_worker);
+//      }
+//    }
+//  }
+//
+//  void ProcessGet(MessagePtr& msg) override {
+//    // 1. Before get: cache faster worker
+//    int worker = Zoo::Get()->rank_to_worker_id(msg->src());
+//    if (worker_add_clocks_->local_clock(worker) >
+//        worker_add_clocks_->global_clock()) {
+//      // Will wait for other worker finished Add
+//      msg_get_cache_.Push(msg);
+//      return;
+//    }
+//    // 2. Process Get
+//    Server::ProcessGet(msg);
+//    // 3. After get: process cached process add if necessary
+//    if (worker_get_clocks_->Update(worker)) {
+//      CHECK(msg_get_cache_.Empty());
+//      while (!msg_add_cache_.Empty()) {
+//        MessagePtr add_msg;
+//        CHECK(msg_add_cache_.TryPop(add_msg));
+//        int add_worker = Zoo::Get()->rank_to_worker_id(add_msg->src());
+//        Server::ProcessAdd(add_msg);
+//        worker_add_clocks_->Update(add_worker);
+//      }
+//    }
+//  }
+//
+//private:
+//  std::unique_ptr<VectorClock> worker_get_clocks_;
+//  std::unique_ptr<VectorClock> worker_add_clocks_;
+//
+//  MtQueue<MessagePtr> msg_add_cache_;
+//  MtQueue<MessagePtr> msg_get_cache_;
+//};
+
+class WithBackupSyncServer : public Server {
 public:
-  SyncServer() : Server() {
-    int num_worker = Zoo::Get()->num_workers();
-    worker_get_clocks_.reset(new VectorClock(num_worker));
-    worker_add_clocks_.reset(new VectorClock(num_worker));
+  WithBackupSyncServer() : Server() {
+    num_worker_ = Zoo::Get()->num_workers();
+    double backup_ratio = (double)MV_CONFIG_backup_worker_ratio / 100;
+    num_sync_worker_ = num_worker_ - 
+                       static_cast<int>(backup_ratio * num_worker_);
+    CHECK(num_sync_worker_ > 0 && num_sync_worker_ <= num_worker_);
+    if (num_sync_worker_ == num_worker_) {
+      Log::Info("No backup worker, using the sync mode\n");
+    }
+    Log::Info("Sync with backup worker start: num_sync_worker = %d," 
+              "num_total_worker = %d\n", num_sync_worker_, num_worker_);
+    worker_get_clocks_.reset(new VectorClock(num_worker_));
+    worker_add_clocks_.reset(new VectorClock(
+      num_worker_, num_worker_ - num_sync_worker_));
  }

  // make some modification to suit to the sync server
  // please not use in other place, may different with the general vector clock
  class VectorClock {
  public:
-    explicit VectorClock(int n) : local_clock_(n, 0), global_clock_(0) {}
+    VectorClock(int num_worker, int num_backup_worker = 0) :
+      local_clock_(num_worker, 0), global_clock_(0), num_worker_(num_worker), 
+      num_sync_worker_(num_worker - num_backup_worker), progress_(0) {}

-    bool Update(int i) {
-      ++local_clock_[i];
-      if (global_clock_ < *(std::min_element(std::begin(local_clock_),
-        std::end(local_clock_)))) {
+    // Return true when global clock meet the sync condition
+    // sync: all worker reach the same clock
+    // backup-worker-sync: sync-workers reach the same clock
+    virtual bool Update(int i) {
+      if (local_clock_[i]++ == global_clock_) {
+        ++progress_;
+      }
+      if (progress_ >= num_sync_worker_) {
        ++global_clock_;
+        progress_ = 0;
+        for (auto i : local_clock_) {
+          if (i > global_clock_) ++progress_;
+        }
        if (global_clock_ == *(std::max_element(std::begin(local_clock_),
          std::end(local_clock_)))) {
          return true;
@ -97,21 +222,27 @@ public:
    int local_clock(int i) const { return local_clock_[i]; }
    int global_clock() const { return global_clock_; }

-  private:
+  protected:
    std::vector<int> local_clock_;
    int global_clock_;
+    int num_worker_;
+    int num_sync_worker_;
+    int progress_;
  };
 protected:
  void ProcessAdd(MessagePtr& msg) override {
    // 1. Before add: cache faster worker
    int worker = Zoo::Get()->rank_to_worker_id(msg->src());
    if (worker_get_clocks_->local_clock(worker) >
-        worker_get_clocks_->global_clock()) {
+      worker_get_clocks_->global_clock()) {
      msg_add_cache_.Push(msg);
      return;
    }
    // 2. Process Add
-    Server::ProcessAdd(msg);
+    if (worker_add_clocks_->local_clock(worker) >= 
+        worker_add_clocks_->global_clock()) {
+      Server::ProcessAdd(msg);
+    }
    // 3. After add: process cached process get if necessary
    if (worker_add_clocks_->Update(worker)) {
      CHECK(msg_add_cache_.Empty());
@ -129,7 +260,7 @@ protected:
    // 1. Before get: cache faster worker
    int worker = Zoo::Get()->rank_to_worker_id(msg->src());
    if (worker_add_clocks_->local_clock(worker) >
-        worker_add_clocks_->global_clock()) {
+      worker_add_clocks_->global_clock()) {
      // Will wait for other worker finished Add
      msg_get_cache_.Push(msg);
      return;
@ -143,7 +274,10 @@ protected:
        MessagePtr add_msg;
        CHECK(msg_add_cache_.TryPop(add_msg));
        int add_worker = Zoo::Get()->rank_to_worker_id(add_msg->src());
-        Server::ProcessAdd(add_msg);
+        if (worker_add_clocks_->local_clock(add_worker) >=
+          worker_add_clocks_->global_clock()) {
+          Server::ProcessAdd(msg);
+        };
        worker_add_clocks_->Update(add_worker);
      }
    }
@ -155,11 +289,21 @@ private:

  MtQueue<MessagePtr> msg_add_cache_;
  MtQueue<MessagePtr> msg_get_cache_;
+
+  // num_worker_ - num_sync_worker_ = num_backup_worker_
+  int num_sync_worker_;
+  int num_worker_;
 };

 Server* Server::GetServer() {
-  if (MV_CONFIG_sync) return new SyncServer();
-  return new Server();
+  if (!MV_CONFIG_sync) {
+    Log::Info("Create a async server\n");
+    return new Server();
+  }
+  // if (MV_CONFIG_backup_worker_ratio > 0.0) {
+  Log::Info("Create a sync server\n");
+  return new WithBackupSyncServer();
+  // }
 }

 }  // namespace multiverso
--- a/src/table.cpp
+++ b/src/table.cpp
@ -1,13 +1,17 @@
 #include "multiverso/table_interface.h"
+
+#include <mutex>
+
+#include "multiverso/dashboard.h"
+#include "multiverso/updater/updater.h"
 #include "multiverso/util/log.h"
 #include "multiverso/util/waiter.h"
 #include "multiverso/zoo.h"
-#include "multiverso/dashboard.h"
-#include "multiverso/updater/updater.h"

 namespace multiverso {

 WorkerTable::WorkerTable() {
+  msg_id_ = 0;
  table_id_ = Zoo::Get()->RegisterTable(this);
 }

@ -30,8 +34,10 @@ void WorkerTable::Add(Blob keys, Blob values,
 }

 int WorkerTable::GetAsync(Blob keys) {
+  m_.lock();
  int id = msg_id_++;
-  waitings_[id] = new Waiter();
+  waitings_.push_back(new Waiter());
+  m_.unlock();
  MessagePtr msg(new Message());
  msg->set_src(Zoo::Get()->rank());
  msg->set_type(MsgType::Request_Get);
@ -44,8 +50,10 @@ int WorkerTable::GetAsync(Blob keys) {

 int WorkerTable::AddAsync(Blob keys, Blob values,
                          const UpdateOption* option) {
+  m_.lock();
  int id = msg_id_++;
-  waitings_[id] = new Waiter();
+  waitings_.push_back(new Waiter());
+  m_.unlock();
  MessagePtr msg(new Message());
  msg->set_src(Zoo::Get()->rank());
  msg->set_type(MsgType::Request_Add);
@ -63,21 +71,32 @@ int WorkerTable::AddAsync(Blob keys, Blob values,
 }

 void WorkerTable::Wait(int id) {
-  CHECK(waitings_.find(id) != waitings_.end());
+  // CHECK(waitings_.find(id) != waitings_.end());
+  m_.lock();
  CHECK(waitings_[id] != nullptr);
-  waitings_[id]->Wait();
+  Waiter* w = waitings_[id];
+  m_.unlock();
+
+  w->Wait();
+
+  m_.lock();
  delete waitings_[id];
  waitings_[id] = nullptr;
+  m_.unlock();
 }

 void WorkerTable::Reset(int msg_id, int num_wait) {
+  m_.lock();
  CHECK_NOTNULL(waitings_[msg_id]);
  waitings_[msg_id]->Reset(num_wait);
+  m_.unlock();
 }

 void WorkerTable::Notify(int id) {
+  m_.lock();
  CHECK_NOTNULL(waitings_[id]);
  waitings_[id]->Notify();
+  m_.unlock();
 }

 WorkerTable* TableHelper::CreateTable() {