WebRTC中音視頻服務質量QoS之RTT衡量網絡往返時延的加權平均RTT計算機制?詳解

WebRTC中音視頻服務質量QoS之RTT衡量網絡往返時延加權平均RTT計算機制?的詳解

WebRTC中音視頻服務質量QoS之RTT衡量網絡往返時延加權平均RTT計算機制?的詳解
前言
一、 RTT 網絡往返時延的原理?
- 1、基于發送端（SR/RR 模式）
- - ①. ?基本定義?
  - ②. ?計算 RTT 網絡往返時延的原理?
  - ③ 發送 Sender Report (SR) 協議
  - - SenderReport 協議的格式
    - 組織SR協議
    - SR和RR中都有ReportBlock數據塊保存 LSR和DLSR的信息
    - SR和RR中都有ReportBlock協議解析
  - ④ 發送ReceiverReport（RR）協議
  - - ReceiverReport協議格式
    - 組織 ReceiverReport(RR)數據
    - 終止計算rtt往返時延加權平均RTT計算機制?
    - - 定時計算 WebRTC中默認1秒
- 2、基于接收端（RTCP XR 模式）
- - 觸發條件?：接收端僅拉流（不發送媒體數據），通過 ?RTCP Extended Reports (XR)? 擴展協議實現 RTT 探測?
二、網絡質量評估算法之時延加權平均RTT計算機制?
三、 rtp和rtcp發送包列表數據保存時間（WebRTC根據rtt計算的）

WebRTC專題開嗨鴨！！！

一、 WebRTC 線程模型

1、WebRTC中線程模型和常見線程模型介紹

2、WebRTC網絡PhysicalSocketServer之WSAEventselect模型使用

二、 WebRTC媒體協商

1、WebRTC媒體協商之SDP中JsepSessionDescription類結構分析

2、WebRTC媒體協商之CreatePeerConnectionFactory、CreatePeerConnection、CreateOffer

3、WebRTC之證書(certificate)生成的時機分析

4、WebRTC源碼之RtpTransceiver添加視頻軌道的AddTrack函數中橋接模式的流程分析

三、 WebRTC 音頻數據采集

1、WebRTC源碼之音頻設備播放流程源碼分析

2、WebRTC源碼之音頻設備的錄制流程源碼分析

四、 WebRTC 音頻引擎(編解碼和3A算法)

五、 WebRTC 視頻數據采集

六、 WebRTC 視頻引擎( 編解碼)

七、 WebRTC 網絡傳輸

1、WebRTC的ICE之STUN協議

2、WebRTC的ICE之Dtls/SSL/TLSv1.x協議詳解

八、 WebRTC服務質量(Qos)

1、WebRTC中RTCP協議詳解

2、WebRTC中RTP協議詳解

3、WebRTC之NACK、RTX 在什么時機判斷丟包發送NACK請求和RTX丟包重傳

4、WebRTC源碼之視頻質量統計數據的數據結構分析

5、WebRTC源碼之RTCPReceiver源碼分析

6、WebRTC中音視頻服務質量QoS之RTT衡量網絡往返時延加權平均RTT計算機制?的詳解

九、 NetEQ

十、 Simulcast與SVC

前言

一、 RTT 網絡往返時延的原理?

WebRTC 提供 ?兩種 RTT 計算模式?，適應不同傳輸場景

1、基于發送端（SR/RR 模式）

*** 觸發條件?：發送端周期性發送 ?Sender Report (SR)?，接收端回應 ?Receiver Report (RR)?? ***

①. ?基本定義?

	?DLSR? 表示自接收端最后一次收到發送端 Sender Report (SR) 到生成當前 Receiver Report (RR) 的時間間隔，單位為 ?1/65536 秒??1。若接收端未收到過 SR 報文，則 DLSR 值為零?1。

②. ?計算 RTT 網絡往返時延的原理?

	在端到端通信中（以端點 A 和 B 為例）：?A 發送 SR?：記錄發送時間 t1（即 LSR，Last SR Timestamp）?2。?B 接收 SR?：記錄接收時間 last_recv_time?2。?B 發送 RR?：計算從 last_recv_time 到當前時間的延遲（即 DLSR），并附加到 RR 報文?2。?A 接收 RR?：根據公式 RTT = 當前時間 - LSR - DLSR 計算往返時間。

公式： ${RTT=T_{current} ? T_ {LSR} ? \frac{T_{DLSR}}{65536}}$ (單位：秒)

參數說明?：

?
$T_ {LSR}$ ：發送端最后一次 SR 的 NTP 時間戳（中間 32 位）?3。
? $T_{DLSR?}$ ：接收端處理 SR 到生成 RR 的延遲（單位：1/65536 秒）?

③ 發送 Sender Report (SR) 協議

SenderReport 協議的格式


//    Sender report (SR) (RFC 3550).
//     0                   1                   2                   3
//     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
//    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
//    |V=2|P|    RC   |   PT=SR=200   |             length            |
//    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
//  0 |                         SSRC of sender                        |
//    +=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+
//  4 |              NTP timestamp, most significant word             |
//    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
//  8 |             NTP timestamp, least significant word             |
//    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
// 12 |                         RTP timestamp                         |
//    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
// 16 |                     sender's packet count                     |
//    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
// 20 |                      sender's octet count                     |
// 24 +=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+

組織SR協議


std::unique_ptr<rtcp::RtcpPacket> RTCPSender::BuildSR(const RtcpContext& ctx) {// Timestamp shouldn't be estimated before first media frame.RTC_DCHECK_GE(last_frame_capture_time_ms_, 0);// The timestamp of this RTCP packet should be estimated as the timestamp of// the frame being captured at this moment. We are calculating that// timestamp as the last frame's timestamp + the time since the last frame// was captured.int rtp_rate = rtp_clock_rates_khz_[last_payload_type_];if (rtp_rate <= 0) {rtp_rate =(audio_ ? kBogusRtpRateForAudioRtcp : kVideoPayloadTypeFrequency) /1000;}// Round now_us_ to the closest millisecond, because Ntp time is rounded// when converted to milliseconds,uint32_t rtp_timestamp =timestamp_offset_ + last_rtp_timestamp_ +((ctx.now_us_ + 500) / 1000 - last_frame_capture_time_ms_) * rtp_rate;rtcp::SenderReport* report = new rtcp::SenderReport();report->SetSenderSsrc(ssrc_);report->SetNtp(TimeMicrosToNtp(ctx.now_us_));report->SetRtpTimestamp(rtp_timestamp);report->SetPacketCount(ctx.feedback_state_.packets_sent);report->SetOctetCount(ctx.feedback_state_.media_bytes_sent);// TODO@chensong  2025-03-15  獲取當前發送 report->SetReportBlocks(CreateReportBlocks(ctx.feedback_state_));return std::unique_ptr<rtcp::RtcpPacket>(report);
}

SR和RR中都有ReportBlock數據塊保存 LSR和DLSR的信息


// From RFC 3550, RTP: A Transport Protocol for Real-Time Applications.
//
// RTCP report block (RFC 3550).
//
//     0                   1                   2                   3
//     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
//    +=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+
//  0 |                 SSRC_1 (SSRC of first source)                 |
//    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
//  4 | fraction lost |       cumulative number of packets lost       |
//    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
//  8 |           extended highest sequence number received           |
//    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
// 12 |                      interarrival jitter                      |
//    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
// 16 |                         last SR (LSR)                         |
//    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
// 20 |                   delay since last SR (DLSR)                  |
// 24 +=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+

SR和RR中都有ReportBlock協議解析

last_sr_ ：發送端發送時間

delay_since_last_sr_ : 是遠端最后接受SR或者RR包的時間

bool ReportBlock::Parse(const uint8_t* buffer, size_t length) 
{RTC_DCHECK(buffer != nullptr);if (length < ReportBlock::kLength){RTC_LOG(LS_ERROR) << "Report Block should be 24 bytes long";return false;}// 接收到的媒體源ssrcsource_ssrc_ = ByteReader<uint32_t>::ReadBigEndian(&buffer[0]);// TODO@chensong 2022-10-19  丟包率 fraction_lost/**TODO@chensong 2023-03-07  某時刻收到的有序包的數量Count = transmitted-retransmitte,當前時刻為Count2,上一時刻為Count1;接收端以一定的頻率發送RTCP包（RR、REMB、NACK等）時，會統計兩次發送間隔之間(fraction)的接收包信息。接收端發送的RR包中包含兩個丟包:一個是fraction_lost，是兩次統計間隔間的丟包率(以256為基數換算成8bit)。一個是cumulative number of packets lost，是總的累積丟包。 **/fraction_lost_ = buffer[4];// 接收開始丟包總數， 遲到包不算丟包，重傳有可以導致負數cumulative_lost_ = ByteReader<int32_t, 3>::ReadBigEndian(&buffer[5]);// 低16位表示收到的最大seq，高16位表示seq循環次數extended_high_seq_num_ = ByteReader<uint32_t>::ReadBigEndian(&buffer[8]);// rtp包到達時間間隔的統計方差jitter_ = ByteReader<uint32_t>::ReadBigEndian(&buffer[12]);// ntp時間戳的中間32位last_sr_ = ByteReader<uint32_t>::ReadBigEndian(&buffer[16]);// 記錄上一個接收SR的時間與上一個發送SR的時間差delay_since_last_sr_ = ByteReader<uint32_t>::ReadBigEndian(&buffer[20]);return true;
}

④ 發送ReceiverReport（RR）協議

ReceiverReport協議格式


// RTCP receiver report (RFC 3550).
//
//   0                   1                   2                   3
//   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
//  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
//  |V=2|P|    RC   |   PT=RR=201   |             length            |
//  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
//  |                     SSRC of packet sender                     |
//  +=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+
//  |                         report block(s)                       |
//  |                            ....                               |

組織 ReceiverReport(RR)數據

在RTCPSender類中BuildRR方法中調用 GetFeedbackState方法獲取 ReportBlock數據

調用流程

RTCPSender類BuildRR —> ModuleRtpRtcpImpl::GetFeedbackState獲取 remote_sender_rtp_time_(遠端發送時間)和 last_received_sr_ntp_ （最后一次接受時間）
—>LastReceivedNTP 方法調用NTP方法
–>RTCPReceiver類NTP 獲取 remote_sender_rtp_time_(遠端發送時間)和 last_received_sr_ntp_ （最后一次接受時間）


std::unique_ptr<rtcp::RtcpPacket> RTCPSender::BuildRR(const RtcpContext& ctx) {rtcp::ReceiverReport* report = new rtcp::ReceiverReport();report->SetSenderSsrc(ssrc_);// TODO@chensong 2025-03-15  rtp_rtcp_impl.cc ->  ModuleRtpRtcpImpl::GetFeedbackStatereport->SetReportBlocks(CreateReportBlocks(ctx.feedback_state_));return std::unique_ptr<rtcp::RtcpPacket>(report);
}// TODO(pbos): Handle media and RTX streams separately (separate RTCP
// feedbacks).
RTCPSender::FeedbackState ModuleRtpRtcpImpl::GetFeedbackState() {RTCPSender::FeedbackState state;// This is called also when receiver_only is true. Hence below// checks that rtp_sender_ exists.if (rtp_sender_) {StreamDataCounters rtp_stats;StreamDataCounters rtx_stats;rtp_sender_->GetDataCounters(&rtp_stats, &rtx_stats);state.packets_sent =rtp_stats.transmitted.packets + rtx_stats.transmitted.packets;state.media_bytes_sent = rtp_stats.transmitted.payload_bytes +rtx_stats.transmitted.payload_bytes;state.send_bitrate = rtp_sender_->BitrateSent();}state.module = this;// TODO@chensong 2025-03-15 獲取遠端發送信息包時間 和當前最后接收一包記錄時間LastReceivedNTP(&state.last_rr_ntp_secs, &state.last_rr_ntp_frac,&state.remote_sr);state.last_xr_rtis = rtcp_receiver_.ConsumeReceivedXrReferenceTimeInfo();return state;
}bool RTCPReceiver::NTP(uint32_t* received_ntp_secs,uint32_t* received_ntp_frac,uint32_t* rtcp_arrival_time_secs,uint32_t* rtcp_arrival_time_frac,uint32_t* rtcp_timestamp) const {rtc::CritScope lock(&rtcp_receiver_lock_);if (!last_received_sr_ntp_.Valid()) {return false;}//   TODO@chensong 2025-03-15  last_rr_ntp_frac 發送時間戳// NTP from incoming SenderReport.if (received_ntp_secs) {*received_ntp_secs = remote_sender_ntp_time_.seconds();}if (received_ntp_frac) {*received_ntp_frac = remote_sender_ntp_time_.fractions();}// Rtp time from incoming SenderReport.// TODO@chensong 2025-03-15 遠端接受最后一個rtp包的時間if (rtcp_timestamp) {*rtcp_timestamp = remote_sender_rtp_time_;}// Local NTP time when we received a RTCP packet with a send block.// TODO@chensong 2025-03-15 本地接受最后一個rtcp包的時間if (rtcp_arrival_time_secs) {*rtcp_arrival_time_secs = last_received_sr_ntp_.seconds();}if (rtcp_arrival_time_frac) {*rtcp_arrival_time_frac = last_received_sr_ntp_.fractions();}return true;
}
// 接收SenderReport包信息
void RTCPReceiver::HandleSenderReport(const CommonHeader& rtcp_block,PacketInformation* packet_information) {rtcp::SenderReport sender_report;if (!sender_report.Parse(rtcp_block)) {++num_skipped_packets_;return;}const uint32_t remote_ssrc = sender_report.sender_ssrc();packet_information->remote_ssrc = remote_ssrc;UpdateTmmbrRemoteIsAlive(remote_ssrc);// Have I received RTP packets from this party?if (remote_ssrc_ == remote_ssrc) {// Only signal that we have received a SR when we accept one.packet_information->packet_type_flags |= kRtcpSr;// TODO@chensong 2025-03-15   SR => RR remote_sender_ntp_time_ = sender_report.ntp();remote_sender_rtp_time_ = sender_report.rtp_timestamp();last_received_sr_ntp_ = TimeMicrosToNtp(clock_->TimeInMicroseconds());} else {// We will only store the send report from one source, but// we will store all the receive blocks.packet_information->packet_type_flags |= kRtcpRr;}for (const rtcp::ReportBlock& report_block : sender_report.report_blocks()) {HandleReportBlock(report_block, packet_information, remote_ssrc);}
}

終止計算rtt往返時延加權平均RTT計算機制?

定時計算 WebRTC中默認1秒

在ModuleRtpRtcpImpl類中Process方法中統計加權平均RTT計算機制?


// Process any pending tasks such as timeouts (non time critical events).
void ModuleRtpRtcpImpl::Process() {const int64_t now = clock_->TimeInMilliseconds();next_process_time_ = now + kRtpRtcpMaxIdleTimeProcessMs;if (rtp_sender_) {if (now >= last_bitrate_process_time_ + kRtpRtcpBitrateProcessTimeMs) {rtp_sender_->ProcessBitrate();last_bitrate_process_time_ = now;next_process_time_ =std::min(next_process_time_, now + kRtpRtcpBitrateProcessTimeMs);}}bool process_rtt = now >= last_rtt_process_time_ + kRtpRtcpRttProcessTimeMs;if (rtcp_sender_.Sending()) {// Process RTT if we have received a report block and we haven't// processed RTT for at least |kRtpRtcpRttProcessTimeMs| milliseconds.if (rtcp_receiver_.LastReceivedReportBlockMs() > last_rtt_process_time_ &&process_rtt) {std::vector<RTCPReportBlock> receive_blocks;rtcp_receiver_.StatisticsReceived(&receive_blocks);int64_t max_rtt = 0;for (std::vector<RTCPReportBlock>::iterator it = receive_blocks.begin();it != receive_blocks.end(); ++it) {int64_t rtt = 0;rtcp_receiver_.RTT(it->sender_ssrc, &rtt, NULL, NULL, NULL);max_rtt = (rtt > max_rtt) ? rtt : max_rtt;}// Report the rtt.if (rtt_stats_ && max_rtt != 0)rtt_stats_->OnRttUpdate(max_rtt);}// Verify receiver reports are delivered and the reported sequence number// is increasing.if (rtcp_receiver_.RtcpRrTimeout()) {RTC_LOG_F(LS_WARNING) << "Timeout: No RTCP RR received.";} else if (rtcp_receiver_.RtcpRrSequenceNumberTimeout()) {RTC_LOG_F(LS_WARNING) << "Timeout: No increase in RTCP RR extended ""highest sequence number.";}if (remote_bitrate_ && rtcp_sender_.TMMBR()) {unsigned int target_bitrate = 0;std::vector<unsigned int> ssrcs;if (remote_bitrate_->LatestEstimate(&ssrcs, &target_bitrate)) {if (!ssrcs.empty()) {target_bitrate = target_bitrate / ssrcs.size();}rtcp_sender_.SetTargetBitrate(target_bitrate);}}} else {// Report rtt from receiver.if (process_rtt) {int64_t rtt_ms;if (rtt_stats_ && rtcp_receiver_.GetAndResetXrRrRtt(&rtt_ms)) {rtt_stats_->OnRttUpdate(rtt_ms);}}}// Get processed rtt.if (process_rtt) {last_rtt_process_time_ = now;next_process_time_ = std::min(next_process_time_, last_rtt_process_time_ + kRtpRtcpRttProcessTimeMs);if (rtt_stats_) {// TODO@chensong 2025-03-15  1秒更新一次 rtt    公式/*TODO@chensong 2025-03-15 加權平均RTT計算機制?在實時通信場景（如WebRTC）中，RTT（往返時延）的平滑計算對網絡狀態感知和擁塞控制至關重要。通過 ?加權移動平均（Weighted Moving Average）? 對RTT值進行動態調整，可有效平衡歷史數據與實時測量值的影響，抑制短期波動帶來的干擾。以下是核心實現邏輯：?1. 公式定義??計算方式?：新平均RTT由 ?歷史平均值（old_avg）? 與 ?最新測量值（new_sample）? 按權重合成，公式為：textCopy Codeavg_rtt = 0.7 * old_avg + 0.3 * new_sample  其中，歷史數據權重為70%（0.7），新樣本權重為30%（0.3）?23。?數學意義?：?舊值主導（70%）?：確保長期趨勢穩定，避免偶發延遲突變（如網絡抖動）對整體估計的過度影響?23。?新值補充（30%）?：快速響應網絡狀態的漸進變化（如帶寬增減或路由切換）?*/// Make sure we have a valid RTT before setting.int64_t last_rtt = rtt_stats_->LastProcessedRtt();if (last_rtt >= 0)set_rtt_ms(last_rtt);}}if (rtcp_sender_.TimeToSendRTCPReport())rtcp_sender_.SendRTCP(GetFeedbackState(), kRtcpReport);if (TMMBR() && rtcp_receiver_.UpdateTmmbrTimers()) {rtcp_receiver_.NotifyTmmbrUpdated();}
}

2、基于接收端（RTCP XR 模式）

觸發條件?：接收端僅拉流（不發送媒體數據），通過 ?RTCP Extended Reports (XR)? 擴展協議實現 RTT 探測?

實現步驟?：

網關發送 ?RRTR 報文?（含 NTP 時間戳 $T_{RRTR}$ )
接收端回復 ?DLRR 報文?，包含
- 原 $T_{RRTR}$ (即為LRR)
- 處理延遲 $T_{DLSR}$ (接收 RRTR 到發送 DLRR 的時間)
網關計算公式

$RTT = {T_{current}}$ - ${T_{TRR}}$ - ${T_{DLSR}}$

二、網絡質量評估算法之時延加權平均RTT計算機制?

加權平均RTT計算機制?
在實時通信場景（如WebRTC）中，RTT（往返時延）的平滑計算對網絡狀態感知和擁塞控制至關重要。通過 ?加權移動平均（Weighted Moving Average）?
對RTT值進行動態調整，可有效平衡歷史數據與實時測量值的影響，抑制短期波動帶來的干擾。以下是核心實現邏輯：

?1. 公式定義??計算方式?：新平均RTT由 ?歷史平均值（old_avg）? 與 ?最新測量值（new_sample）? 按權重合成，公式為：avg_rtt = 0.7 * old_avg + 0.3 * new_sample  其中，歷史數據權重為70%（0.7），新樣本權重為30%（0.3）?23。?數學意義?：?舊值主導（70%）?：確保長期趨勢穩定，避免偶發延遲突變（如網絡抖動）對整體估計的過度影響?23。?新值補充（30%）?：快速響應網絡狀態的漸進變化（如帶寬增減或路由切換）?

三、 rtp和rtcp發送包列表數據保存時間（WebRTC根據rtt計算的）

void RtpPacketHistory::CullOldPackets(int64_t now_ms) 
{//TODO@chensong 2025-03-15 比如NACK（否定確認）或ARQ（自動重傳請求）中的緩沖區管理策略有關。//  根據 rtt 放棄 rtp包 // 公式 ： 淘汰時間 = 3 × max(基準時間, 3 × 當前RTT)// 基準時間通常為 1000ms（兜底值，防止 RTT 過小導致緩存不足）int64_t packet_duration_ms = std::max(kMinPacketDurationRtt * rtt_ms_, kMinPacketDurationMs);while (!packet_history_.empty()){auto stored_packet_it = packet_history_.find(*start_seqno_);RTC_DCHECK(stored_packet_it != packet_history_.end());if (packet_history_.size() >= kMaxCapacity /* 9600*/) {// We have reached the absolute max capacity, remove one packet// unconditionally.RemovePacket(stored_packet_it);continue;}const StoredPacket& stored_packet = stored_packet_it->second;if (!stored_packet.send_time_ms) {// Don't remove packets that have not been sent.return;}if (*stored_packet.send_time_ms + packet_duration_ms > now_ms) {// Don't cull packets too early to avoid failed retransmission requests.return;}if (packet_history_.size() >= number_to_store_ ||(mode_ == StorageMode::kStoreAndCull && *stored_packet.send_time_ms + (packet_duration_ms * kPacketCullingDelayFactor) <= now_ms)) {// Too many packets in history, or this packet has timed out. Remove it// and continue.RemovePacket(stored_packet_it);}else {// No more packets can be removed right now.return;}}
}