netty四种BUFFER的内存测试

分别使用下面四种buffer测试

ByteBufoutBuf=UnpooledByteBufAllocator.DEFAULT.heapBuffer();

//ByteBufoutBuf=PooledByteBufAllocator.DEFAULT.heapBuffer();

//ByteBufoutBuf=PooledByteBufAllocator.DEFAULT.buffer();

//ByteBufoutBuf=UnpooledByteBufAllocator.DEFAULT.buffer();

 

outBuf=outBuf.order(ByteOrder.LITTLE_ENDIAN);

outBuf.writeInt(17);

outBuf.writeInt(112);

outBuf.writeShort(11);

outBuf.writeShort(5);

outBuf.writeBytes(bytes);

ctx.channel().writeAndFlush(outBuf);

 

 

UnpooledByteBufAllocator.DEFAULT.heapBuffer()

吞吐量：30144/秒

PooledByteBufAllocator.DEFAULT.heapBuffer() 

吞吐量：30327/秒

 

 

 

PooledByteBufAllocator.DEFAULT.buffer() 

吞吐量：32082/秒

 

 

 

 

 

UnpooledByteBufAllocator.DEFAULT.buffer()

吞吐量：32292/秒

 

 

可以发现PooledByteBufAllocator.DEFAULT.heapBuffer()情况使用的堆内存最多，为72.7MB，按理说，使用内存池可重用内存，理论上是最少的，看细细研究是这样的结论：

在启动程序时，设置了四个线程，

final DefaultEventExecutorGroupexecutorGroup = new DefaultEventExecutorGroup(4, new DefaultThreadFactory("decode-worker-thread-pool"));

 

netty避免每个线程对内存池的竞争，在每个线程都提供了PoolThreadCache线程内的内存池，

final classPoolThreadCache {

 

   finalPoolArena<byte[]> heapArena;

   finalPoolArena<ByteBuffer>directArena;

 

   // TODO: Test if adding padding helps undercontention

   //private long pad0, pad1, pad2, pad3,pad4, pad5, pad6, pad7;

 

   PoolThreadCache(PoolArena<byte[]> heapArena, PoolArena<ByteBuffer>directArena) {

        this.heapArena = heapArena;

        this.directArena = directArena;

   }

}

 

 

PoolArena里的PoolChunk初始化大小16MB,有四个线程，于是看到的就如图