优化网络IO使用buffer读取

stream/receive_buffer.go

@@ -0,0 +1,36 @@
+package stream
+
+const (
+	ReceiveBufferUdpBlockCount = 200
+
+	ReceiveBufferTCPBlockCount = 100
+)
+
+// ReceiveBuffer 收流缓冲区. 网络收流->解析流->封装流->发送流是同步的,从解析到发送可能耗时,从而影响读取网络流. 使用收流缓冲区,可有效降低出现此问题的概率.
+// 从网络IO读取数据->送给解复用器, 此过程需做到无内存拷贝
+// rtmp和1078推流直接使用ReceiveBuffer
+// 国标推流,UDP收流都要经过jitter buffer处理, 还是需要拷贝一次, 没必要使用ReceiveBuffer. TCP全都使用ReceiveBuffer, 区别在于多端口模式, 第一包传给source, 单端口模式先解析出ssrc, 找到source. 后续再传给source.
+type ReceiveBuffer struct {
+	blockSize  int //单个缓存块大小
+	blockCount int //缓存块数据流. 应当和Source的数据输入管道容量保持一致.
+	data       []byte
+	index      int
+}
+
+func (r *ReceiveBuffer) GetBlock() []byte {
+	bytes := r.data[r.index*r.blockSize:]
+	r.index = r.index + 1%r.blockCount
+	return bytes[:r.blockSize]
+}
+
+func NewReceiveBuffer(blockSize, blockCount int) *ReceiveBuffer {
+	return &ReceiveBuffer{blockSize: blockSize, blockCount: blockCount, data: make([]byte, blockSize*blockCount), index: 0}
+}
+
+func NewUDPReceiveBuffer() *ReceiveBuffer {
+	return NewReceiveBuffer(1500, ReceiveBufferUdpBlockCount)
+}
+
+func NewTCPReceiveBuffer() *ReceiveBuffer {
+	return NewReceiveBuffer(4096*20, ReceiveBufferTCPBlockCount)
+}