Class MpscFFLamportOffHeapFixedSizeRingBuffer


  • public final class MpscFFLamportOffHeapFixedSizeRingBuffer
    extends OffHeapFixedMessageSizeRingBuffer
    A Multi-Producer-Single-Consumer ring buffer. This implies that any thread may call the write methods, but only a single thread may call reads for correctness to maintained.
    This implementation follows patterns for False Sharing protection.
    This implementation is using the Fast Flow method for polling from the queue (with minor change to correctly publish the index) and an extension of the Leslie Lamport concurrent queue algorithm (originated by Martin Thompson) on the producer side.
    • Field Detail

      • consumerIndexCacheAddress

        private final long consumerIndexCacheAddress
    • Constructor Detail

      • MpscFFLamportOffHeapFixedSizeRingBuffer

        public MpscFFLamportOffHeapFixedSizeRingBuffer​(int capacity,
                                                       int primitiveMessageSize,
                                                       int referenceMessageSize)
      • MpscFFLamportOffHeapFixedSizeRingBuffer

        protected MpscFFLamportOffHeapFixedSizeRingBuffer​(java.nio.ByteBuffer buff,
                                                          int capacity,
                                                          boolean isProducer,
                                                          boolean isConsumer,
                                                          boolean initialize,
                                                          int primitiveMessageSize,
                                                          java.lang.Object[] references,
                                                          int referenceMessageSize)
        This is to be used for an IPC queue with the direct buffer used being a memory mapped file.
        Parameters:
        buff -
        capacity -
    • Method Detail

      • slowPathWriteAcquire

        private long slowPathWriteAcquire​(long wrapPoint)
      • writeAcquire

        protected final long writeAcquire()
        Description copied from class: ProxyChannelRingBuffer
        Acquire an offset to write to. If there's no space available a wait strategy may be used.
        Specified by:
        writeAcquire in class OffHeapFixedMessageSizeRingBuffer
        Returns:
        a base address for a message acquired to be written, or EOF if none is available
      • writeRelease

        protected final void writeRelease​(long offset)
        Specified by:
        writeRelease in class OffHeapFixedMessageSizeRingBuffer
        Parameters:
        offset - the base address of a message that we are done writing and can be read now
      • writeRelease

        protected final void writeRelease​(long offset,
                                          int callTypeId)
        Description copied from class: ProxyChannelRingBuffer
        Ordered store of the callTypeId for the message at offset
        Specified by:
        writeRelease in class ProxyChannelRingBuffer
        Parameters:
        offset - the offset that was released for writing
        callTypeId - A unique ID for the call
      • casProducerIndex

        private boolean casProducerIndex​(long expected,
                                         long update)
      • lpConsumerIndexCache

        private long lpConsumerIndexCache()
      • spConsumerIndexCache

        private void spConsumerIndexCache​(long value)
      • spReadReleaseState

        private static void spReadReleaseState​(long offset)