Changeset 917 for XIOS/trunk/src/context_client.cpp

Timestamp:

08/04/16 16:24:20 (8 years ago)

Author:

rlacroix

Message:

Fix the client/server communication protocol.

In the some extreme cases a deadlock could occur. To fix this, the number of buffered events must be properly limited.

If you noticed decreased performance due to this commit, please let us know about it.

Fixes ticket #91.

File:

• XIOS/trunk/src/context_client.cpp (modified) (4 diffs)

XIOS/trunk/src/context_client.cpp

@@ -21 +21 @@
     */
     CContextClient::CContextClient(CContext* parent, MPI_Comm intraComm_, MPI_Comm interComm_, CContext* cxtSer)
-     : mapBufferSize_(), parentServer(cxtSer)
+     : mapBufferSize_(), parentServer(cxtSer), maxBufferedEvents(4)
     {
       context = parent;
@@ -70 +70 @@
 
       timeLine = 0;
-
     }
 
@@ -179 +178 @@
         mapBufferSize_[rank] = CXios::minBufferSize;
       }
-      CClientBuffer* buffer = buffers[rank] = new CClientBuffer(interComm, rank, mapBufferSize_[rank]);
+      CClientBuffer* buffer = buffers[rank] = new CClientBuffer(interComm, rank, mapBufferSize_[rank], maxBufferedEvents);
       // Notify the server
       CBufferOut* bufOut = buffer->getBuffer(sizeof(StdSize));
@@ -219 +218 @@
 
    /*!
-   Set buffer size for each connection
-   \param [in] mapSize mapping rank of connected server to size of allocated buffer
-   */
-   void CContextClient::setBufferSize(const std::map<int,StdSize>& mapSize)
+    * Set the buffer size for each connection. Warning: This function is collective.
+    *
+    * \param [in] mapSize maps the rank of the connected servers to the size of the correspoinding buffer
+    * \param [in] maxEventSize maps the rank of the connected servers to the size of the biggest event
+   */
+   void CContextClient::setBufferSize(const std::map<int,StdSize>& mapSize, const std::map<int,StdSize>& maxEventSize)
    {
      mapBufferSize_ = mapSize;
+
+     // Compute the maximum number of events that can be safely buffered.
+     double minBufferSizeEventSizeRatio = std::numeric_limits<double>::max();
+     for (std::map<int,StdSize>::const_iterator it = mapSize.begin(), ite = mapSize.end(); it != ite; ++it)
+     {
+       double ratio = double(it->second) / maxEventSize.at(it->first);
+       if (ratio < minBufferSizeEventSizeRatio) minBufferSizeEventSizeRatio = ratio;
+     }
+     MPI_Allreduce(MPI_IN_PLACE, &minBufferSizeEventSizeRatio, 1, MPI_DOUBLE, MPI_MIN, intraComm);
+
+     if (minBufferSizeEventSizeRatio < 1.0)
+       ERROR("void CContextClient::setBufferSize(const std::map<int,StdSize>& mapSize, const std::map<int,StdSize>& maxEventSize)",
+             << "The buffer sizes and the maximum events sizes are incoherent.");
+
+     maxBufferedEvents = size_t(2 * minBufferSizeEventSizeRatio) // there is room for two local buffers on the server
+                          + size_t(minBufferSizeEventSizeRatio)  // one local buffer can always be fully used
+                          + 1;                                   // the other local buffer might contain only one event
    }