implement tcp bbr on ns3 (在ns3上实现TCP BBR)

 After several days of work and debug, I implement TCP BBR algorithms on ns3.33.
 One thing is to get instant rtt instead of smothed rtt.
 The second is I notice some difference in tcp-rate-ops.cc (ns3) by comparison with tcp_rate.c (linux tcp stack).

const TcpRateOps::TcpRateSample &
TcpRateLinux::GenerateSample (uint32_t delivered, uint32_t lost, bool is_sack_reneg,
                              uint32_t priorInFlight, const Time &minRtt){
    if (m_rateSample.m_priorTime == Seconds (0) || is_sack_reneg)
    {
      NS_LOG_INFO ("PriorTime is zero, invalidating sample");
      m_rateSample.m_delivered = -1;
      m_rateSample.m_interval = Seconds (0);
      m_rateSampleTrace (m_rateSample);
      return m_rateSample;
    }
  if (m_rateSample.m_interval < minRtt)
    {
      NS_LOG_INFO ("Sampling interval is invalid");
      m_rateSample.m_interval  = Seconds (0);
      m_rateSample.m_priorTime = Seconds (0); // To make rate sample invalid
      m_rateSampleTrace (m_rateSample);
      return m_rateSample;
    }                              
 }

 When m_priorTime is reset as zero in TcpRateSample, m_rateSample.m_ackElapsed will sample a time intervel equal to Simulator::Now (). And m_rateSample.m_priorTime will be update and the acked packet is valid (m_rateSample.m_priorTime!=Seconds(0)) in GenerateSample. A quite low bandwidth value will be sampled. Such low bandwidth will be used by long term bandwidth sampling in BBR algorithm. The flow will gain lower bandwidth here after and weird behavior will be observed.

void
TcpRateLinux::SkbDelivered (TcpTxItem * skb){
    if (m_rateSample.m_priorDelivered == 0
      || skbInfo.m_delivered > m_rateSample.m_priorDelivered)
    {
      m_rateSample.m_ackElapsed       = Simulator::Now () - m_rateSample.m_priorTime;
      m_rateSample.m_priorDelivered   = skbInfo.m_delivered;
      m_rateSample.m_priorTime        = skbInfo.m_deliveredTime;
      m_rateSample.m_isAppLimited     = skbInfo.m_isAppLimited;
      m_rateSample.m_sendElapsed      = skb->GetLastSent () - skbInfo.m_firstSent;

      m_rateSampleTrace (m_rateSample);

      m_rate.m_firstSentTime          = skb->GetLastSent ();
    }
}

 So change is made on SkbDelivered function:

  if (m_rateSample.m_priorDelivered == 0
      || skbInfo.m_delivered > m_rateSample.m_priorDelivered)
    {
      m_rateSample.m_priorDelivered   = skbInfo.m_delivered;
      m_rateSample.m_priorTime        = skbInfo.m_deliveredTime;
      m_rateSample.m_isAppLimited     = skbInfo.m_isAppLimited;
      m_rateSample.m_sendElapsed      = skb->GetLastSent () - skbInfo.m_firstSent;
      m_rate.m_firstSentTime          = skb->GetLastSent ();
    }
    m_rateSample.m_ackElapsed        = Simulator::Now () - m_rateSample.m_priorTime;
    m_rateSampleTrace (m_rateSample);

The third change is in tcp_socket_base.cc

uint32_t
TcpSocketBase::SendDataPacket (SequenceNumber32 seq, uint32_t maxSize, bool withAck){
	m_rateOps->SkbSent(outItem, isStartOfTransmission&&(m_tcb->m_highTxMark==m_tcb->m_nextTxSequence));
}

 I test three TCP BBR flows on a 6Mbps, 50 one way propagation delay p2p link.

Result:

 Estimated bandwidth at sender:

 Taces rtt of each flow:

  The code is uploaded to github [6].
Reference:
[1] Reproduce-TCP-BBR-in-ns-3
[2] linux bbr
[3] ns-3-dev-git
[4] test quic bbr on ns3
[5] evaluation BBRv2 on ns3
[6] tcp-bbr-ns3 code