diff --git a/CHANGES.html b/CHANGES.html
index bf927dc3a..09646ab32 100644
--- a/CHANGES.html
+++ b/CHANGES.html
@@ -57,6 +57,7 @@ us a note on ns-developers mailing list.</p>
 <li>A new TCP congestion control, <b>TcpLinuxReno</b>, has been added.</li>
 <li>Added, to <b>PIE queue disc</b>, <b>queue delay calculation using timestamp</b> feature (Linux default behavior), <b>cap drop adjustment</b> feature (Section 5.5 of RFC 8033), <b>ECN</b> (Section 5.1 of RFC 8033) and <b>derandomization</b> feature (Section 5.4 of RFC 8033).</li>
 <li>Added <b>L4S Mode</b> to FqCoDel and CoDel queue discs</li>
+<li> A model for <b>dynamic pacing</b> has been added to TCP.</li>
 </ul>
 <h2>Changes to existing API:</h2>
 <ul>
@@ -73,8 +74,9 @@ by including flow-monitor.h you will need to change that to stats-module.h.
 <h2>Changed behavior:</h2>
 <ul>
 <li>Support for <b>RIFS</b> has been dropped from wifi. RIFS has been obsoleted by the 802.11 standard and support for it was not implemented according to the standard.</li>
-<li>The behavior of <b>TcpPrrRecovery</b> algorithm was aligned to that of Linux.</b>
+<li>The behavior of <b>TcpPrrRecovery</b> algorithm was aligned to that of Linux.</li>
 <li> <b>PIE</b> queue disc now uses <b>Timestamp</b> for queue delay calculation as default instead of <b>Dequeue Rate Estimator</b></li>
+<li>TCP pacing, when enabled, now adjusts the rate dynamically based on the window size, rather than just enforcing a constant rate.</li>
 </ul>
 
 <hr>
diff --git a/RELEASE_NOTES b/RELEASE_NOTES
index 21e923bae..2e3539973 100644
--- a/RELEASE_NOTES
+++ b/RELEASE_NOTES
@@ -22,6 +22,7 @@ New user-visible features
   (Section 5.5 of RFC 8033), ECN (Section 5.1 of RFC 8033) and derandomization
   feature (Section 5.4 of RFC 8033).
 - (traffic-control) Add support for L4S mode to CoDel and FqCoDel queue discs
+- (tcp) Support for dynamic pacing in TCP.
 
 Bugs fixed
 ----------
diff --git a/examples/tcp/tcp-pacing.cc b/examples/tcp/tcp-pacing.cc
index 6a148f2a7..969a23643 100644
--- a/examples/tcp/tcp-pacing.cc
+++ b/examples/tcp/tcp-pacing.cc
@@ -1,6 +1,6 @@
 /* -*- Mode:C++; c-file-style:"gnu"; indent-tabs-mode:nil; -*- */
 /*
- * Copyright (c) 2017 NITK Surathkal
+ * Copyright (c) 2020 NITK Surathkal
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License version 2 as
@@ -15,21 +15,72 @@
  * along with this program; if not, write to the Free Software
  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  *
- * Author: Vivek Jain <jain.vivek.anand@gmail.com>
+ * Authors: Vivek Jain <jain.vivek.anand@gmail.com>
+ *          Deepak Kumaraswamy <deepakkavoor99@gmail.com>
  */
 
-// Network topology
+// The following network topology is used in this example, and is taken from
+// Figure 2 of https://homes.cs.washington.edu/~tom/pubs/pacing.pdf
+//
+//    n0                          n4
+//    |                           |
+//    |(4x Mbps, 5ms)             |(4x Mbps, 5ms)
+//    |                           |
+//    |                           |
+//    |      (x Mbps, 40ms)       |
+//    n2 ------------------------ n3
+//    |                           |
+//    |                           |
+//    |(4x Mbps, 5ms)             |(4x Mbps, 5ms)
+//    |                           |
+//    n1                          n5
+//
 //
-//       n0 ----------- n1
-//            40 Gbps
-//            0.01 ms
 
-// This programs illustrates how TCP pacing can be used and how user can set
-// pacing rate. The program gives information about each flow like transmitted
-// and received bytes (packets) and throughput of that flow. Currently, it is 
-// using TCP NewReno but in future after having congestion control algorithms 
-// which can change pacing rate can be used.
+// This example illustrates how TCP pacing can be enabled on a socket.
+// Two long-running TCP flows are instantiated at nodes n0 and n1 to
+// send data over a bottleneck link (n2->n3) to sink nodes n4 and n5.
+// At the end of the simulation, the IP-level flow monitor tool will
+// print out summary statistics of the flows.  The flow monitor detects
+// four flows, but that is because the flow records are unidirectional;
+// the latter two flows reported are actually ack streams.
+//
+// At the end of this simulation, data files are also generated
+// that track changes in Congestion Window, Slow Start threshold and
+// TCP pacing rate for the first flow (n0). Additionally, a data file
+// that contains information about packet transmission and reception times
+// (collected through TxTrace and RxTrace respectively) is also produced.
+// This transmission and reception (ack) trace is the most direct way to
+// observe the effects of pacing.  All the above information is traced
+// just for the single node n0.
+//
+// A small amount of randomness is introduced to the program to control
+// the start time of the flows.
+//
+// This example has pacing enabled by default, which means that TCP
+// does not send packets back-to-back, but instead paces them out over
+// an RTT. The size of initial congestion window is set to 10, and pacing
+// of the initial window is enabled. The available command-line options and
+// their default values can be observed in the usual way by running the
+// program to print the help info; i.e.: ./waf --run 'tcp-pacing --PrintHelp'
+//
+// When pacing is disabled, TCP sends eligible packets back-to-back. The
+// differences in behaviour when pacing is disabled can be observed from the
+// packet transmission data file. For instance, one can observe that
+// packets in the initial window are sent one after the other simultaneously,
+// without any inter-packet gaps. Another instance is when n0 receives a
+// packet in the form of an acknowledgement, and sends out data packets without
+// pacing them.
+//
+// Although this example serves as a useful demonstration of how pacing could
+// be enabled/disabled in ns-3 TCP congestion controls, we could not observe
+// significant improvements in throughput for the above topology when pacing
+// was enabled. In future, one could try and incorporate models such as
+// TCP Prague and ACK-filtering, which may show a stronger performance
+// impact for TCP pacing.
 
+#include <iomanip>
+#include <iostream>
 #include <string>
 #include <fstream>
 #include "ns3/core-module.h"
@@ -39,100 +90,219 @@
 #include "ns3/network-module.h"
 #include "ns3/packet-sink.h"
 #include "ns3/flow-monitor-module.h"
+#include "ns3/ipv4-global-routing-helper.h"
+#include "ns3/traffic-control-module.h"
 
 using namespace ns3;
 
 NS_LOG_COMPONENT_DEFINE ("TcpPacingExample");
 
+std::ofstream cwndStream;
+std::ofstream pacingRateStream;
+std::ofstream ssThreshStream;
+std::ofstream packetTraceStream;
+
+static void
+CwndTracer (uint32_t oldval, uint32_t newval)
+{
+  cwndStream << std::fixed << std::setprecision (6) << Simulator::Now ().GetSeconds () << std::setw (12) << newval << std::endl;
+}
+
+static void
+PacingRateTracer (DataRate oldval, DataRate newval)
+{
+  pacingRateStream << std::fixed << std::setprecision (6) << Simulator::Now ().GetSeconds () << std::setw (12) << newval.GetBitRate () / 1e6 << std::endl;
+}
+
+static void
+SsThreshTracer (uint32_t oldval, uint32_t newval)
+{
+  ssThreshStream << std::fixed << std::setprecision (6) << Simulator::Now ().GetSeconds () << std::setw (12) << newval << std::endl;
+}
+
+static void
+TxTracer (Ptr<const Packet> p, Ptr<Ipv4> ipv4, uint32_t interface)
+{
+  packetTraceStream << std::fixed << std::setprecision (6) << Simulator::Now ().GetSeconds () << " tx " << p->GetSize () << std::endl;
+}
+
+static void
+RxTracer (Ptr<const Packet> p, Ptr<Ipv4> ipv4, uint32_t interface)
+{
+  packetTraceStream << std::fixed << std::setprecision (6) << Simulator::Now ().GetSeconds () << " rx " << p->GetSize () << std::endl;
+}
+
+void
+ConnectSocketTraces (void)
+{
+  Config::ConnectWithoutContext ("/NodeList/0/$ns3::TcpL4Protocol/SocketList/0/CongestionWindow", MakeCallback (&CwndTracer));
+  Config::ConnectWithoutContext ("/NodeList/0/$ns3::TcpL4Protocol/SocketList/0/PacingRate", MakeCallback (&PacingRateTracer));
+  Config::ConnectWithoutContext ("/NodeList/0/$ns3::TcpL4Protocol/SocketList/0/SlowStartThreshold", MakeCallback (&SsThreshTracer));
+  Config::ConnectWithoutContext ("/NodeList/0/$ns3::Ipv4L3Protocol/Tx", MakeCallback (&TxTracer));
+  Config::ConnectWithoutContext ("/NodeList/0/$ns3::Ipv4L3Protocol/Rx", MakeCallback (&RxTracer));
+}
+
 int
 main (int argc, char *argv[])
 {
-
   bool tracing = false;
-  uint32_t maxBytes = 0;
-  uint32_t TCPFlows = 1;
+
+  uint32_t maxBytes = 0; // value of zero corresponds to unlimited send
+  std::string transportProtocol = "ns3::TcpNewReno";
+
+  Time simulationEndTime = Seconds (5);
+  DataRate bottleneckBandwidth ("10Mbps");  // value of x as shown in the above network topology
+  Time bottleneckDelay = MilliSeconds (40);
+  DataRate regLinkBandwidth = DataRate (4 * bottleneckBandwidth.GetBitRate ());
+  Time regLinkDelay = MilliSeconds (5);
+  DataRate maxPacingRate ("4Gbps");
+
   bool isPacingEnabled = true;
-  std::string pacingRate = "4Gbps";
-  bool isSack = false;
-  uint32_t maxPackets = 0;
+  bool useEcn = true;
+  bool useQueueDisc = true;
+  bool shouldPaceInitialWindow = true;
+
+  // Configure defaults that are not based on explicit command-line arguments
+  // They may be overridden by general attribute configuration of command line
+  Config::SetDefault ("ns3::TcpL4Protocol::SocketType", TypeIdValue (TypeId::LookupByName (transportProtocol)));
+  Config::SetDefault ("ns3::TcpSocket::InitialCwnd", UintegerValue (10));
 
   CommandLine cmd (__FILE__);
-  cmd.AddValue ("tracing", "Flag to enable/disable tracing", tracing);
-  cmd.AddValue ("maxBytes",
-                "Total number of bytes for application to send", maxBytes);
-  cmd.AddValue ("maxPackets",
-                "Total number of bytes for application to send", maxPackets);
-  cmd.AddValue ("TCPFlows", "Number of application flows between sender and receiver", TCPFlows);
-  cmd.AddValue ("Pacing", "Flag to enable/disable pacing in TCP", isPacingEnabled);
-  cmd.AddValue ("Sack", "Flag to enable/disable sack in TCP", isSack);
-  cmd.AddValue ("PacingRate", "Max Pacing Rate in bps", pacingRate);
+  cmd.AddValue ("tracing", "Flag to enable/disable Ascii and Pcap tracing", tracing);
+  cmd.AddValue ("maxBytes", "Total number of bytes for application to send", maxBytes);
+  cmd.AddValue ("isPacingEnabled", "Flag to enable/disable pacing in TCP", isPacingEnabled);
+  cmd.AddValue ("maxPacingRate", "Max Pacing Rate", maxPacingRate);
+  cmd.AddValue ("useEcn", "Flag to enable/disable ECN", useEcn);
+  cmd.AddValue ("useQueueDisc", "Flag to enable/disable queue disc on bottleneck", useQueueDisc);
+  cmd.AddValue ("shouldPaceInitialWindow", "Flag to enable/disable pacing of TCP initial window", shouldPaceInitialWindow);
+  cmd.AddValue ("simulationEndTime", "Simulation end time", simulationEndTime);
   cmd.Parse (argc, argv);
 
-  if (maxPackets != 0 )
-    {
-      maxBytes = 500 * maxPackets;
-    }
-
-  Config::SetDefault ("ns3::TcpSocketState::MaxPacingRate", StringValue (pacingRate));
+  // Configure defaults based on command-line arguments
   Config::SetDefault ("ns3::TcpSocketState::EnablePacing", BooleanValue (isPacingEnabled));
-  Config::SetDefault ("ns3::TcpSocketBase::Sack", BooleanValue (isSack));
+  Config::SetDefault ("ns3::TcpSocketState::PaceInitialWindow", BooleanValue (shouldPaceInitialWindow));
+  Config::SetDefault ("ns3::TcpSocketBase::UseEcn", (useEcn ? EnumValue (TcpSocketState::On) : EnumValue (TcpSocketState::Off)));
+  Config::SetDefault ("ns3::TcpSocketState::MaxPacingRate", DataRateValue (maxPacingRate));
 
   NS_LOG_INFO ("Create nodes.");
-  NodeContainer nodes;
-  nodes.Create (2);
+  NodeContainer c;
+  c.Create (6);
 
   NS_LOG_INFO ("Create channels.");
-  PointToPointHelper pointToPoint;
-  pointToPoint.SetDeviceAttribute ("DataRate", StringValue ("40Gbps"));
-  pointToPoint.SetChannelAttribute ("Delay", StringValue ("0.01ms"));
+  NodeContainer n0n2 = NodeContainer (c.Get (0), c.Get (2));
+  NodeContainer n1n2 = NodeContainer (c.Get (1), c.Get (2));
 
-  NetDeviceContainer devices;
-  devices = pointToPoint.Install (nodes);
+  NodeContainer n2n3 = NodeContainer (c.Get (2), c.Get (3));
 
-  InternetStackHelper internet;
-  internet.Install (nodes);
+  NodeContainer n3n4 = NodeContainer (c.Get (3), c.Get (4));
+  NodeContainer n3n5 = NodeContainer (c.Get (3), c.Get (5));
+
+  //Define Node link properties
+  PointToPointHelper regLink;
+  regLink.SetDeviceAttribute ("DataRate", DataRateValue (regLinkBandwidth));
+  regLink.SetChannelAttribute ("Delay", TimeValue (regLinkDelay));
+
+  NetDeviceContainer d0d2 = regLink.Install (n0n2);
+  NetDeviceContainer d1d2 = regLink.Install (n1n2);
+  NetDeviceContainer d3d4 = regLink.Install (n3n4);
+  NetDeviceContainer d3d5 = regLink.Install (n3n5);
+
+  PointToPointHelper bottleNeckLink;
+  bottleNeckLink.SetDeviceAttribute ("DataRate", DataRateValue (bottleneckBandwidth));
+  bottleNeckLink.SetChannelAttribute ("Delay", TimeValue (bottleneckDelay));
+
+  NetDeviceContainer d2d3 = bottleNeckLink.Install (n2n3);
+
+  //Install Internet stack
+  InternetStackHelper stack;
+  stack.Install (c);
+
+  // Install traffic control
+  if (useQueueDisc)
+    {
+      TrafficControlHelper tchBottleneck;
+      tchBottleneck.SetRootQueueDisc ("ns3::FqCoDelQueueDisc");
+      tchBottleneck.Install (d2d3);
+    }
 
   NS_LOG_INFO ("Assign IP Addresses.");
   Ipv4AddressHelper ipv4;
   ipv4.SetBase ("10.1.1.0", "255.255.255.0");
-  Ipv4InterfaceContainer i = ipv4.Assign (devices);
+  Ipv4InterfaceContainer regLinkInterface0 = ipv4.Assign (d0d2);
+
+  ipv4.SetBase ("10.1.2.0", "255.255.255.0");
+  Ipv4InterfaceContainer regLinkInterface1 = ipv4.Assign (d1d2);
+
+  ipv4.SetBase ("10.1.3.0", "255.255.255.0");
+  Ipv4InterfaceContainer bottleneckInterface = ipv4.Assign (d2d3);
+
+  ipv4.SetBase ("10.1.4.0", "255.255.255.0");
+  Ipv4InterfaceContainer regLinkInterface4 = ipv4.Assign (d3d4);
+
+  ipv4.SetBase ("10.1.5.0", "255.255.255.0");
+  Ipv4InterfaceContainer regLinkInterface5 = ipv4.Assign (d3d5);
+
+  Ipv4GlobalRoutingHelper::PopulateRoutingTables ();
 
   NS_LOG_INFO ("Create Applications.");
 
-  ApplicationContainer sourceApps;
-  ApplicationContainer sinkApps;
-  for (uint32_t iterator = 0; iterator < TCPFlows; iterator++)
-    {
-      uint16_t port = 10000 + iterator;
+  // Two Sink Applications at n4 and n5
+  uint16_t sinkPort = 8080;
+  Address sinkAddress4 (InetSocketAddress (regLinkInterface4.GetAddress (1), sinkPort)); // interface of n4
+  Address sinkAddress5 (InetSocketAddress (regLinkInterface5.GetAddress (1), sinkPort)); // interface of n5
+  PacketSinkHelper packetSinkHelper ("ns3::TcpSocketFactory", InetSocketAddress (Ipv4Address::GetAny (), sinkPort));
+  ApplicationContainer sinkApps4 = packetSinkHelper.Install (c.Get (4)); //n4 as sink
+  ApplicationContainer sinkApps5 = packetSinkHelper.Install (c.Get (5)); //n5 as sink
 
-      BulkSendHelper source  ("ns3::TcpSocketFactory",
-                              InetSocketAddress (i.GetAddress (1), port));
-      // Set the amount of data to send in bytes.  Zero is unlimited.
-      source.SetAttribute ("MaxBytes", UintegerValue (maxBytes));
-      sourceApps.Add (source.Install (nodes.Get (0)));
+  sinkApps4.Start (Seconds (0));
+  sinkApps4.Stop (simulationEndTime);
+  sinkApps5.Start (Seconds (0));
+  sinkApps5.Stop (simulationEndTime);
 
-      PacketSinkHelper sink ("ns3::TcpSocketFactory",
-                             InetSocketAddress (Ipv4Address::GetAny (), port));
-      sinkApps.Add (sink.Install (nodes.Get (1)));
-    }
+  // Randomize the start time between 0 and 1ms
+  Ptr<UniformRandomVariable> uniformRv = CreateObject<UniformRandomVariable> ();
+  uniformRv->SetStream (0);
 
-  sinkApps.Start (Seconds (0.0));
-  sinkApps.Stop (Seconds (5));
-  sourceApps.Start (Seconds (1));
-  sourceApps.Stop (Seconds (5));
+  // Two Source Applications at n0 and n1
+  BulkSendHelper source0 ("ns3::TcpSocketFactory", sinkAddress4);
+  BulkSendHelper source1 ("ns3::TcpSocketFactory", sinkAddress5);
+  // Set the amount of data to send in bytes.  Zero is unlimited.
+  source0.SetAttribute ("MaxBytes", UintegerValue (maxBytes));
+  source1.SetAttribute ("MaxBytes", UintegerValue (maxBytes));
+  ApplicationContainer sourceApps0 = source0.Install (c.Get (0));
+  ApplicationContainer sourceApps1 = source1.Install (c.Get (1));
+
+  sourceApps0.Start (MicroSeconds (uniformRv->GetInteger (0, 1000)));
+  sourceApps0.Stop (simulationEndTime);
+  sourceApps1.Start (MicroSeconds (uniformRv->GetInteger (0, 1000)));
+  sourceApps1.Stop (simulationEndTime);
 
   if (tracing)
     {
       AsciiTraceHelper ascii;
-      pointToPoint.EnableAsciiAll (ascii.CreateFileStream ("tcp-pacing.tr"));
-      pointToPoint.EnablePcapAll ("tcp-pacing", false);
+      regLink.EnableAsciiAll (ascii.CreateFileStream ("tcp-dynamic-pacing.tr"));
+      regLink.EnablePcapAll ("tcp-dynamic-pacing", false);
     }
 
+  cwndStream.open ("tcp-dynamic-pacing-cwnd.dat", std::ios::out);
+  cwndStream << "#Time(s) Congestion Window (B)" << std::endl;
+
+  pacingRateStream.open ("tcp-dynamic-pacing-pacing-rate.dat", std::ios::out);
+  pacingRateStream << "#Time(s) Pacing Rate (Mb/s)" << std::endl;
+
+  ssThreshStream.open ("tcp-dynamic-pacing-ssthresh.dat", std::ios::out);
+  ssThreshStream << "#Time(s) Slow Start threshold (B)" << std::endl;
+
+  packetTraceStream.open ("tcp-dynamic-pacing-packet-trace.dat", std::ios::out);
+  packetTraceStream << "#Time(s) tx/rx size (B)" << std::endl;
+
+  Simulator::Schedule (MicroSeconds (1001), &ConnectSocketTraces);
+
   FlowMonitorHelper flowmon;
   Ptr<FlowMonitor> monitor = flowmon.InstallAll ();
 
   NS_LOG_INFO ("Run Simulation.");
-  Simulator::Stop (Seconds (5));
+  Simulator::Stop (simulationEndTime);
   Simulator::Run ();
 
   monitor->CheckForLostPackets ();
@@ -141,19 +311,19 @@ main (int argc, char *argv[])
   for (std::map<FlowId, FlowMonitor::FlowStats>::const_iterator i = stats.begin (); i != stats.end (); ++i)
     {
       Ipv4FlowClassifier::FiveTuple t = classifier->FindFlow (i->first);
-      if (t.sourceAddress == "10.1.1.2")
-        {
-          continue;
-        }
+
       std::cout << "Flow " << i->first  << " (" << t.sourceAddress << " -> " << t.destinationAddress << ")\n";
       std::cout << "  Tx Packets: " << i->second.txPackets << "\n";
       std::cout << "  Tx Bytes:   " << i->second.txBytes << "\n";
-      std::cout << "  TxOffered:  " << i->second.txBytes * 8.0 / 9.0 / 1000 / 1000  << " Mbps\n";
+      std::cout << "  TxOffered:  " << i->second.txBytes * 8.0 / simulationEndTime.GetSeconds () / 1000 / 1000  << " Mbps\n";
       std::cout << "  Rx Packets: " << i->second.rxPackets << "\n";
       std::cout << "  Rx Bytes:   " << i->second.rxBytes << "\n";
-      std::cout << "  Throughput: " << i->second.rxBytes * 8.0 / 9.0 / 1000 / 1000  << " Mbps\n";
+      std::cout << "  Throughput: " << i->second.rxBytes * 8.0 / simulationEndTime.GetSeconds () / 1000 / 1000  << " Mbps\n";
     }
 
+
+  cwndStream.close ();
+  pacingRateStream.close ();
+  ssThreshStream.close ();
   Simulator::Destroy ();
-  NS_LOG_INFO ("Done.");
 }
diff --git a/src/internet/doc/tcp.rst b/src/internet/doc/tcp.rst
index 09ed0188f..b217760ff 100644
--- a/src/internet/doc/tcp.rst
+++ b/src/internet/doc/tcp.rst
@@ -968,10 +968,11 @@ environment. Some differences were noted:
 * Linux maintains its congestion window in segments and not bytes, and
   the arithmetic is not floating point, so some differences in the
   evolution of congestion window have been observed.
-* Linux uses pacing, while ns-3 currently does not provide a dynamically
-  adjusting pacing implementation; segments are sent out at the line rate
-  unless the user has enabled pacing and set the maximum pacing rate to
-  less than the line rate.
+* Linux uses pacing, where packets to be sent are paced out at regular
+  intervals. However, if at any instant the number of segments that can 
+  be sent are less than two, Linux does not pace them and instead sends 
+  them back-to-back. Currently, ns-3 paces out all packets eligible to 
+  be sent in the same manner.
 * Linux implements a state called 'Congestion Window Reduced' (CWR) 
   immediately following a cwnd reduction, and performs proportional rate
   reduction similar to how a fast retransmit event is handled.  During
@@ -1150,6 +1151,81 @@ The following issues are yet to be addressed:
    outgoing TCP sessions (e.g. a TCP may perform ECN echoing but not set the
    ECT codepoints on its outbound data segments).
 
+Support for Dynamic Pacing
+++++++++++++++++++++++++++
+
+TCP pacing refers to the sender-side practice of scheduling the transmission
+of a burst of eligible TCP segments across a time interval such as
+a TCP RTT, to avoid or reduce bursts.  Historically,
+TCP used the natural ACK clocking mechanism to pace segments, but some
+network paths introduce aggregation (bursts of ACKs arriving) or ACK
+thinning, either of which disrupts ACK clocking.
+Some latency-sensitive congestion controls under development (Prague, BBR)
+require pacing to operate effectively.
+
+Until recently, the state of the art in Linux was to support pacing in one
+of two ways:
+
+1) fq/pacing with sch_fq
+2) TCP internal pacing
+
+The presentation by Dumazet and Cheng at IETF 88 summarizes:
+https://www.ietf.org/proceedings/88/slides/slides-88-tcpm-9.pdf
+
+The first option was most often used when offloading (TSO) was enabled and
+when the sch_fq scheduler was used at the traffic control (qdisc) sublayer.  In
+this case, TCP was responsible for setting the socket pacing rate, but
+the qdisc sublayer would enforce it.  When TSO was enabled, the kernel
+would break a large burst into smaller chunks, with dynamic sizing based
+on the pacing rate, and hand off the segments to the fq qdisc for
+pacing.
+
+The second option was used if sch_fq was not enabled; TCP would be
+responsible for internally pacing.
+
+In 2018, Linux switched to an Early Departure Model (EDM): https://lwn.net/Articles/766564/.
+
+TCP pacing in Linux was added in kernel 3.12, and authors chose to allow
+a pacing rate of 200% against the current rate, to allow probing for
+optimal throughput even during slow start phase.  Some refinements were
+added in https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=43e122b014c9, 
+in which Google reported that it was better to apply
+a different ratio (120%) in Congestion Avoidance phase.  Furthermore,
+authors found that after cwnd reduction, it was helpful to become more
+conservative and switch to the conservative ratio (120%) as soon as 
+cwnd >= ssthresh/2, as the initial ramp up (when ssthresh is infinite) still
+allows doubling cwnd every other RTT.  Linux also does not pace the initial
+window (IW), typically 10 segments in practice.
+
+Linux has also been observed to not pace if the number of eligible segments
+to be sent is exactly two; they will be sent back to back.  If three or
+more, the first two are sent immediately, and additional segments are paced
+at the current pacing rate.     
+
+In ns-3, the model is as follows.  There is no TSO/sch_fq model; only
+internal pacing according to current Linux policy.
+
+Pacing may be enabled for any TCP congestion control, and a maximum
+pacing rate can be set.  Furthermore, dynamic pacing is enabled for
+all TCP variants, according to the following guidelines.
+
+* Pacing of the initial window (IW) is not done by default but can be
+  separately enabled.
+
+* Pacing of the initial slow start, after IW, is done according to the
+  pacing rate of 200% of the current rate, to allow for window growth
+  This pacing rate can be configured to a different value than 200%.
+
+* Pacing of congestion avoidance phase is done at a pacing rate of 120% of
+  current rate.  This can be configured to a different value than 120%.
+
+* Pacing of subsequent slow start is done according to the following
+  heuristic.  If cwnd < ssthresh/2, such as after a timeout or idle period,
+  pace at the slow start rate (200%).  Otherwise, pace at the congestion
+  avoidance rate.
+
+Dynamic pacing is demonstrated by the example program ``examples/tcp/tcp-pacing.cc``. 
+
 Validation
 ++++++++++
 
@@ -1188,6 +1264,7 @@ section below on :ref:`Writing-tcp-tests`.
 * **tcp-zero-window-test:** Unit test persist behavior for zero window conditions
 * **tcp-close-test:** Unit test on the socket closing: both receiver and sender have to close their socket when all bytes are transferred
 * **tcp-ecn-test:** Unit tests on Explicit Congestion Notification
+* **tcp-pacing-test:** Unit tests on dynamic TCP pacing rate
 
 Several tests have dependencies outside of the ``internet`` module, so they
 are located in a system test directory called ``src/test/ns3tcp``. Three
diff --git a/src/internet/model/tcp-socket-base.cc b/src/internet/model/tcp-socket-base.cc
index 7900253f9..ceae971e2 100644
--- a/src/internet/model/tcp-socket-base.cc
+++ b/src/internet/model/tcp-socket-base.cc
@@ -196,6 +196,10 @@ TcpSocketBase::GetTypeId (void)
                      "Highest ack received from peer",
                      MakeTraceSourceAccessor (&TcpSocketBase::m_highRxAckMark),
                      "ns3::TracedValueCallback::SequenceNumber32")
+    .AddTraceSource ("PacingRate",
+                     "The current TCP pacing rate",
+                     MakeTraceSourceAccessor (&TcpSocketBase::m_pacingRateTrace),
+                     "ns3::TracedValueCallback::DataRate")
     .AddTraceSource ("CongestionWindow",
                      "The TCP connection's congestion window",
                      MakeTraceSourceAccessor (&TcpSocketBase::m_cWndTrace),
@@ -249,13 +253,17 @@ TcpSocketBase::TcpSocketBase (void)
 
   m_tcb->m_rxBuffer = CreateObject<TcpRxBuffer> ();
 
-  m_tcb->m_currentPacingRate = m_tcb->m_maxPacingRate;
+  m_tcb->m_pacingRate = m_tcb->m_maxPacingRate;
   m_pacingTimer.SetFunction (&TcpSocketBase::NotifyPacingPerformed, this);
 
   m_tcb->m_sendEmptyPacketCallback = MakeCallback (&TcpSocketBase::SendEmptyPacket, this);
 
   bool ok;
 
+  ok = m_tcb->TraceConnectWithoutContext ("PacingRate",
+                                          MakeCallback (&TcpSocketBase::UpdatePacingRateTrace, this));
+  NS_ASSERT (ok == true);
+
   ok = m_tcb->TraceConnectWithoutContext ("CongestionWindow",
                                           MakeCallback (&TcpSocketBase::UpdateCwnd, this));
   NS_ASSERT (ok == true);
@@ -365,7 +373,7 @@ TcpSocketBase::TcpSocketBase (const TcpSocketBase& sock)
   m_tcb = CopyObject (sock.m_tcb);
   m_tcb->m_rxBuffer = CopyObject (sock.m_tcb->m_rxBuffer);
 
-  m_tcb->m_currentPacingRate = m_tcb->m_maxPacingRate;
+  m_tcb->m_pacingRate = m_tcb->m_maxPacingRate;
   m_pacingTimer.SetFunction (&TcpSocketBase::NotifyPacingPerformed, this);
 
   if (sock.m_congestionControl)
@@ -387,6 +395,9 @@ TcpSocketBase::TcpSocketBase (const TcpSocketBase& sock)
 
   bool ok;
 
+  ok = m_tcb->TraceConnectWithoutContext ("PacingRate",
+                                          MakeCallback (&TcpSocketBase::UpdatePacingRateTrace, this));
+
   ok = m_tcb->TraceConnectWithoutContext ("CongestionWindow",
                                           MakeCallback (&TcpSocketBase::UpdateCwnd, this));
   NS_ASSERT (ok == true);
@@ -2092,6 +2103,11 @@ TcpSocketBase::ProcessAck(const SequenceNumber32 &ackNumber, bool scoreboardUpda
             }
         }
     }
+  // Update the pacing rate, since m_congestionControl->IncreaseWindow() or
+  // m_congestionControl->PktsAcked () may change m_tcb->m_cWnd
+  // Make sure that control reaches the end of this function and there is no
+  // return in between
+  UpdatePacingRate ();
 }
 
 /* Received a packet upon LISTEN state. */
@@ -2183,6 +2199,8 @@ TcpSocketBase::ProcessSynSent (Ptr<Packet> packet, const TcpHeader& tcpHeader)
       m_tcb->m_rxBuffer->SetNextRxSequence (tcpHeader.GetSequenceNumber () + SequenceNumber32 (1));
       m_tcb->m_highTxMark = ++m_tcb->m_nextTxSequence;
       m_txBuffer->SetHeadSequence (m_tcb->m_nextTxSequence);
+      // Before sending packets, update the pacing rate based on RTT measurement so far 
+      UpdatePacingRate ();
       SendEmptyPacket (TcpHeader::ACK);
 
       /* Check if we received an ECN SYN-ACK packet. Change the ECN state of sender to ECN_IDLE if receiver has sent an ECN SYN-ACK
@@ -2256,6 +2274,8 @@ TcpSocketBase::ProcessSynRcvd (Ptr<Packet> packet, const TcpHeader& tcpHeader,
       m_delAckCount = m_delAckMaxCount;
       NotifyNewConnectionCreated (this, fromAddress);
       ReceivedAck (packet, tcpHeader);
+      // Update the pacing rate based on RTT measurement so far
+      UpdatePacingRate ();
       // As this connection is established, the socket is available to send data now
       if (GetTxAvailable () > 0)
         {
@@ -2967,20 +2987,24 @@ TcpSocketBase::SendDataPacket (SequenceNumber32 seq, uint32_t maxSize, bool with
   // peer when it is not retransmission.
   NS_ASSERT (isRetransmission || ((m_highRxAckMark + SequenceNumber32 (m_rWnd)) >= (seq + SequenceNumber32 (maxSize))));
 
-  if (m_tcb->m_pacing)
+  if (IsPacingEnabled ())
     {
       NS_LOG_INFO ("Pacing is enabled");
       if (m_pacingTimer.IsExpired ())
         {
-          NS_LOG_DEBUG ("Current Pacing Rate " << m_tcb->m_currentPacingRate);
-          NS_LOG_DEBUG ("Timer is in expired state, activate it " << m_tcb->m_currentPacingRate.CalculateBytesTxTime (sz));
-          m_pacingTimer.Schedule (m_tcb->m_currentPacingRate.CalculateBytesTxTime (sz));
+          NS_LOG_DEBUG ("Current Pacing Rate " << m_tcb->m_pacingRate);
+          NS_LOG_DEBUG ("Timer is in expired state, activate it " << m_tcb->m_pacingRate.Get ().CalculateBytesTxTime (sz));
+          m_pacingTimer.Schedule (m_tcb->m_pacingRate.Get ().CalculateBytesTxTime (sz));
         }
       else
         {
           NS_LOG_INFO ("Timer is already in running state");
         }
     }
+  else
+    {
+      NS_LOG_INFO ("Pacing is disabled");
+    }
 
   if (withAck)
     {
@@ -2995,6 +3019,7 @@ TcpSocketBase::SendDataPacket (SequenceNumber32 seq, uint32_t maxSize, bool with
       m_congestionControl->ReduceCwnd (m_tcb);
       m_tcb->m_ssThresh = m_tcb->m_cWnd;
       m_tcb->m_cWndInfl = m_tcb->m_cWnd;
+      UpdatePacingRate ();
       flags |= TcpHeader::CWR;
       m_ecnCWRSeq = seq;
       NS_LOG_DEBUG (TcpSocketState::EcnStateName[m_tcb->m_ecnState] << " -> ECN_CWR_SENT");
@@ -3133,7 +3158,7 @@ TcpSocketBase::SendPendingData (bool withAck)
   // else branch to control silly window syndrome and Nagle)
   while (availableWindow > 0)
     {
-      if (m_tcb->m_pacing)
+      if (IsPacingEnabled ())
         {
           NS_LOG_INFO ("Pacing is enabled");
           if (m_pacingTimer.IsRunning ())
@@ -3228,14 +3253,14 @@ TcpSocketBase::SendPendingData (bool withAck)
                         " size " << sz);
           m_tcb->m_nextTxSequence += sz;
           ++nPacketsSent;
-          if (m_tcb->m_pacing)
+          if (IsPacingEnabled ())
             {
               NS_LOG_INFO ("Pacing is enabled");
               if (m_pacingTimer.IsExpired ())
                 {
-                  NS_LOG_DEBUG ("Current Pacing Rate " << m_tcb->m_currentPacingRate);
-                  NS_LOG_DEBUG ("Timer is in expired state, activate it " << m_tcb->m_currentPacingRate.CalculateBytesTxTime (sz));
-                  m_pacingTimer.Schedule (m_tcb->m_currentPacingRate.CalculateBytesTxTime (sz));
+                  NS_LOG_DEBUG ("Current Pacing Rate " << m_tcb->m_pacingRate);
+                  NS_LOG_DEBUG ("Timer is in expired state, activate it " << m_tcb->m_pacingRate.Get ().CalculateBytesTxTime (sz));
+                  m_pacingTimer.Schedule (m_tcb->m_pacingRate.Get ().CalculateBytesTxTime (sz));
                   break;
                 }
             }
@@ -4265,6 +4290,12 @@ TcpSocketBase::SetRetxThresh (uint32_t retxThresh)
   m_txBuffer->SetDupAckThresh (retxThresh);
 }
 
+void
+TcpSocketBase::UpdatePacingRateTrace (DataRate oldValue, DataRate newValue)
+{
+  m_pacingRateTrace (oldValue, newValue);
+}
+
 void
 TcpSocketBase::UpdateCwnd (uint32_t oldValue, uint32_t newValue)
 {
@@ -4363,6 +4394,89 @@ TcpSocketBase::NotifyPacingPerformed (void)
   SendPendingData (m_connected);
 }
 
+bool
+TcpSocketBase::IsPacingEnabled (void) const
+{
+  if (!m_tcb->m_pacing)
+    {
+      return false;
+    }
+  else
+    {
+      if (m_tcb->m_paceInitialWindow)
+        {
+          return true;
+        }
+      SequenceNumber32 highTxMark = m_tcb->m_highTxMark; // cast traced value
+      if (highTxMark.GetValue () > (GetInitialCwnd () * m_tcb->m_segmentSize))
+        {
+          return true;
+        }
+    }
+  return false;
+}
+
+void
+TcpSocketBase::UpdatePacingRate (void)
+{
+  NS_LOG_FUNCTION (this << m_tcb);
+
+  // According to Linux, set base pacing rate to (cwnd * mss) / srtt
+  //
+  // In (early) slow start, multiply base by the slow start factor.
+  // In late slow start and congestion avoidance, multiply base by
+  // the congestion avoidance factor.
+  // Comment from Linux code regarding early/late slow start:
+  // Normal Slow Start condition is (tp->snd_cwnd < tp->snd_ssthresh)
+  // If snd_cwnd >= (tp->snd_ssthresh / 2), we are approaching
+  // end of slow start and should slow down.
+
+  // Similar to Linux, do not update pacing rate here if the
+  // congestion control implements TcpCongestionOps::CongControl ()
+  if (m_congestionControl->HasCongControl () || !m_tcb->m_pacing) return;
+
+  double factor;
+  if (m_tcb->m_cWnd < m_tcb->m_ssThresh/2)
+    {
+      NS_LOG_DEBUG ("Pacing according to slow start factor; " << m_tcb->m_cWnd << " " << m_tcb->m_ssThresh);
+      factor = static_cast<double> (m_tcb->m_pacingSsRatio)/100;
+    }
+  else
+    {
+      NS_LOG_DEBUG ("Pacing according to congestion avoidance factor; " << m_tcb->m_cWnd << " " << m_tcb->m_ssThresh);
+      factor = static_cast<double> (m_tcb->m_pacingCaRatio)/100;
+    }
+  Time lastRtt = m_tcb->m_lastRtt.Get (); // Get underlying Time value
+  NS_LOG_DEBUG ("Last RTT is " << lastRtt.GetSeconds ());
+  
+  // Multiply by 8 to convert from bytes per second to bits per second
+  DataRate pacingRate ((std::max (m_tcb->m_cWnd, m_tcb->m_bytesInFlight) * 8 * factor) / lastRtt.GetSeconds ());
+  if (pacingRate < m_tcb->m_maxPacingRate)
+    {
+      NS_LOG_DEBUG ("Pacing rate updated to: " << pacingRate);
+      m_tcb->m_pacingRate = pacingRate;
+    }
+  else
+    {
+      NS_LOG_DEBUG ("Pacing capped by max pacing rate: " << m_tcb->m_maxPacingRate);
+      m_tcb->m_pacingRate = m_tcb->m_maxPacingRate;
+    }
+}
+
+void
+TcpSocketBase::SetPacingStatus (bool pacing)
+{
+  NS_LOG_FUNCTION (this << pacing);
+  m_tcb->m_pacing = pacing;
+}
+
+void
+TcpSocketBase::SetPaceInitialWindow (bool paceWindow)
+{
+  NS_LOG_FUNCTION (this << paceWindow);
+  m_tcb->m_paceInitialWindow = paceWindow;
+}
+
 void
 TcpSocketBase::SetUseEcn (TcpSocketState::UseEcn_t useEcn)
 {
diff --git a/src/internet/model/tcp-socket-base.h b/src/internet/model/tcp-socket-base.h
index ef7a53674..6a4edc78b 100644
--- a/src/internet/model/tcp-socket-base.h
+++ b/src/internet/model/tcp-socket-base.h
@@ -317,6 +317,11 @@ public:
    */
   uint32_t GetRetxThresh (void) const { return m_retxThresh; }
 
+  /**
+   * \brief Callback pointer for pacing rate trace chaining
+   */
+  TracedCallback<DataRate, DataRate>  m_pacingRateTrace;
+
   /**
    * \brief Callback pointer for cWnd trace chaining
    */
@@ -362,6 +367,13 @@ public:
    */
   TracedCallback<Time, Time> m_lastRttTrace;
 
+  /**
+   * \brief Callback function to hook to TcpSocketState pacing rate
+   * \param oldValue old pacing rate value
+   * \param newValue new pacing rate value
+   */
+  void UpdatePacingRateTrace (DataRate oldValue, DataRate newValue);
+
   /**
    * \brief Callback function to hook to TcpSocketState congestion window
    * \param oldValue old cWnd value
@@ -547,6 +559,18 @@ public:
    */
   void SetUseEcn (TcpSocketState::UseEcn_t useEcn);
 
+  /**
+   * \brief Enable or disable pacing
+   * \param pacing Boolean to enable or disable pacing
+   */
+  void SetPacingStatus (bool pacing);
+
+  /**
+   * \brief Enable or disable pacing of the initial window
+   * \param pacing Boolean to enable or disable pacing of the initial window
+   */
+  void SetPaceInitialWindow (bool paceWindow);
+
   // Necessary implementations of null functions from ns3::Socket
   virtual enum SocketErrno GetErrno (void) const;    // returns m_errno
   virtual enum SocketType GetSocketType (void) const; // returns socket type
@@ -1178,6 +1202,19 @@ protected:
    */
   void NotifyPacingPerformed (void);
 
+  /**
+   * \brief Return true if packets in the current window should be paced
+   * \return true if pacing is currently enabled
+   */
+  bool IsPacingEnabled (void) const;
+
+  /**
+   * \brief Dynamically update the pacing rate
+   *
+   * \param tcb internal congestion state
+   */
+  void UpdatePacingRate (void);
+
   /**
    * \brief Add Tags for the Socket
    * \param p Packet
diff --git a/src/internet/model/tcp-socket-state.cc b/src/internet/model/tcp-socket-state.cc
index 63d330b9a..2102a7f69 100644
--- a/src/internet/model/tcp-socket-state.cc
+++ b/src/internet/model/tcp-socket-state.cc
@@ -36,6 +36,22 @@ TcpSocketState::GetTypeId (void)
                    DataRateValue (DataRate ("4Gb/s")),
                    MakeDataRateAccessor (&TcpSocketState::m_maxPacingRate),
                    MakeDataRateChecker ())
+    .AddAttribute ("PacingSsRatio", "Percent pacing rate increase for slow start conditions",
+                   UintegerValue (200),
+                   MakeUintegerAccessor (&TcpSocketState::m_pacingSsRatio),
+                   MakeUintegerChecker<uint16_t> ())
+    .AddAttribute ("PacingCaRatio", "Percent pacing rate increase for congestion avoidance conditions",
+                   UintegerValue (120),
+                   MakeUintegerAccessor (&TcpSocketState::m_pacingCaRatio),
+                   MakeUintegerChecker<uint16_t> ())
+    .AddAttribute ("PaceInitialWindow", "Perform pacing for initial window of data",
+                   BooleanValue (false),
+                   MakeBooleanAccessor (&TcpSocketState::m_paceInitialWindow),
+                   MakeBooleanChecker ())
+    .AddTraceSource ("PacingRate",
+                     "The current TCP pacing rate",
+                     MakeTraceSourceAccessor (&TcpSocketState::m_pacingRate),
+                     "ns3::TracedValueCallback::DataRate")
     .AddTraceSource ("CongestionWindow",
                      "The TCP connection's congestion window",
                      MakeTraceSourceAccessor (&TcpSocketState::m_cWnd),
@@ -92,7 +108,10 @@ TcpSocketState::TcpSocketState (const TcpSocketState &other)
     m_rcvTimestampEchoReply (other.m_rcvTimestampEchoReply),
     m_pacing (other.m_pacing),
     m_maxPacingRate (other.m_maxPacingRate),
-    m_currentPacingRate (other.m_currentPacingRate),
+    m_pacingRate (other.m_pacingRate),
+    m_pacingSsRatio (other.m_pacingSsRatio),
+    m_pacingCaRatio (other.m_pacingCaRatio),
+    m_paceInitialWindow (other.m_paceInitialWindow),
     m_minRtt (other.m_minRtt),
     m_bytesInFlight (other.m_bytesInFlight),
     m_lastRtt (other.m_lastRtt),
diff --git a/src/internet/model/tcp-socket-state.h b/src/internet/model/tcp-socket-state.h
index e649896cf..d400e94b5 100644
--- a/src/internet/model/tcp-socket-state.h
+++ b/src/internet/model/tcp-socket-state.h
@@ -187,7 +187,10 @@ public:
   // Pacing related variables
   bool                   m_pacing            {false}; //!< Pacing status
   DataRate               m_maxPacingRate     {0};    //!< Max Pacing rate
-  DataRate               m_currentPacingRate {0};    //!< Current Pacing rate
+  TracedValue<DataRate>  m_pacingRate {0};           //!< Current Pacing rate
+  uint16_t               m_pacingSsRatio {0};        //!< SS pacing ratio
+  uint16_t               m_pacingCaRatio {0};        //!< CA pacing ratio
+  bool                   m_paceInitialWindow {false}; //!< Enable/Disable pacing for the initial window
 
   Time                   m_minRtt  {Time::Max ()};   //!< Minimum RTT observed throughout the connection
 
diff --git a/src/internet/test/tcp-general-test.cc b/src/internet/test/tcp-general-test.cc
index 9a8043082..af83e9d30 100644
--- a/src/internet/test/tcp-general-test.cc
+++ b/src/internet/test/tcp-general-test.cc
@@ -984,6 +984,40 @@ TcpGeneralTest::SetUseEcn (SocketWho who, TcpSocketState::UseEcn_t useEcn)
     }
 }
 
+void
+TcpGeneralTest::SetPacingStatus (SocketWho who, bool pacing)
+{
+  if (who == SENDER)
+    {
+      m_senderSocket->SetPacingStatus (pacing);
+    }
+  else if (who == RECEIVER)
+    {
+      m_receiverSocket->SetPacingStatus (pacing);
+    }
+  else
+    {
+      NS_FATAL_ERROR ("Not defined");
+    }
+}
+
+void
+TcpGeneralTest::SetPaceInitialWindow (SocketWho who, bool paceWindow)
+{
+  if (who == SENDER)
+    {
+      m_senderSocket->SetPaceInitialWindow (paceWindow);
+    }
+  else if (who == RECEIVER)
+    {
+      m_receiverSocket->SetPaceInitialWindow (paceWindow);
+    }
+  else
+    {
+      NS_FATAL_ERROR ("Not defined");
+    }
+}
+
 void
 TcpGeneralTest::SetInitialSsThresh (SocketWho who, uint32_t initialSsThresh)
 {
diff --git a/src/internet/test/tcp-general-test.h b/src/internet/test/tcp-general-test.h
index dd3bdcfa8..59b070081 100644
--- a/src/internet/test/tcp-general-test.h
+++ b/src/internet/test/tcp-general-test.h
@@ -607,6 +607,22 @@ protected:
    */
   void SetUseEcn (SocketWho who, TcpSocketState::UseEcn_t useEcn);
 
+  /**
+   * \brief Enable or disable pacing in the TCP socket
+   * 
+   * \param who socket
+   * \param pacing Boolean to enable or disable pacing
+   */
+  void SetPacingStatus (SocketWho who, bool pacing);
+
+  /**
+   * \brief Enable or disable pacing of the initial window
+   * 
+   * \param who socket
+   * \param pacing Boolean to enable or disable pacing of initial window
+   */
+  void SetPaceInitialWindow (SocketWho who, bool paceWindow);
+
   /**
    * \brief Forcefully set the initial ssthresh
    *
diff --git a/src/internet/test/tcp-pacing-test.cc b/src/internet/test/tcp-pacing-test.cc
new file mode 100644
index 000000000..900cf9cd0
--- /dev/null
+++ b/src/internet/test/tcp-pacing-test.cc
@@ -0,0 +1,413 @@
+/* -*-  Mode: C++; c-file-style: "gnu"; indent-tabs-mode:nil; -*- */
+/*
+ * Copyright (c) 2020 NITK Surathkal
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation;
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ *
+ * Authors: Deepak Kumaraswamy <deepakkavoor99@gmail.com>
+ *
+ */
+#include "ns3/log.h"
+#include "ns3/simple-channel.h"
+#include "ns3/config.h"
+#include "ns3/test.h"
+#include "tcp-general-test.h"
+
+using namespace ns3;
+
+NS_LOG_COMPONENT_DEFINE ("TcpPacingTestSuite");
+
+/**
+ * \ingroup internet-test
+ * \ingroup tests
+ *
+ * \brief Test the behavior of TCP pacing
+ *
+ * This test checks that packets are paced at correct intervals.  The test
+ * uses a shadow pacing rate calculation assumed to match the internal
+ * pacing calculation.  Therefore, if you modify the values of
+ * pacingSsRatio and pacingCaRatio herein, ensure that you also change
+ * values used in the TCP implementation to match.
+ *
+ * This test environment uses an RTT of 100ms
+ * The time interval between two consecutive packet transmissions is measured
+ * Pacing rate should be at least cwnd / rtt to transmit cwnd segments per
+ * rtt.  Linux multiples this basic ratio by an additional factor, to yield
+ * pacingRate = (factor * cwnd) / rtt
+ * Since pacingRate can also be written as segmentSize / interval
+ * we can solve for interval = (segmentSize * rtt) / (factor * cwnd)
+ *
+ * The test checks whether the measured interval lies within a tolerance
+ * value of the expected interval.  The tolerance or error margin
+ * was chosen to be 10 Nanoseconds, that could be due to delay introduced
+ * by the application's send process.
+ *
+ * This check should not be performed for a packet transmission after the
+ * sender has sent all bytes corresponding to the window and is awaiting an
+ * ACK from the receiver (corresponding to m_isFullCwndSent).
+ * Pacing check should be performed when the sender is actively sending packets from cwnd.
+ *
+ * The same argument applies when the sender has finished sending packets and is awaiting a
+ * FIN/ACK from the receiver, to send the final ACK
+ *
+ * As can be seen in TcpSocketBase::UpdatePacingRate (), different pacing
+ * ratios are used when cwnd < ssThresh / 2 and when cwnd > ssThresh / 2
+ *
+ * A few key points to note:
+ * - In TcpSocketBase, pacing rate is updated whenever an ACK is received.
+ *
+ * - The factors that could contribute to a different value of pacing rate include
+ * congestion window and RTT.
+ *
+ * - However, the cWnd trace is called after Rx () trace and we therefore
+ * update the expected interval in cWnd trace.
+ *
+ * - An RTT trace is also necessary, since using delayed ACKs may lead to a
+ * higher RTT measurement at the sender's end. The expected interval should
+ * be updated here as well.
+ *
+ * - When sending the current packet, TcpSocketBase automatically decides the time at
+ * which next packet should be sent. So, if say packet 3 is sent and an ACK is
+ * received before packet 4 is sent (thus updating the pacing rate), this does not change
+ * the time at which packet 4 is to be sent. When packet 4 is indeed sent later, the new
+ * pacing rate is used to decide when packet 5 will be sent. This behavior is captured in
+ * m_nextPacketInterval, which is not affected by change of pacing rate before the next
+ * packet is sent. The important observation here is to realize the contrast between
+ * m_expectedInterval and m_nextPacketInterval.
+ *
+ */
+class
+TcpPacingTest : public TcpGeneralTest
+{
+public:
+  /**
+   * \brief Constructor.
+   * \param segmentSize Segment size at the TCP layer (bytes).
+   * \param packetSize Size of packets sent at the application layer (bytes).
+   * \param packets Number of packets.
+   * \param pacingSsRatio Pacing Ratio during Slow Start (multiplied by 100)
+   * \param pacingCaRatio Pacing Ratio during Congestion Avoidance (multiplied by 100)
+   * \param ssThresh slow start threshold (bytes)
+   * \param paceInitialWindow whether to pace the initial window
+   * \param delAckMaxCount Delayed ACK max count parameter
+   * \param congControl Type of congestion control.
+   * \param desc The test description.
+   */
+  TcpPacingTest (uint32_t segmentSize, uint32_t packetSize,
+                 uint32_t packets, uint16_t pacingSsRatio, uint16_t pacingCaRatio,
+                 uint32_t ssThresh, bool paceInitialWindow, uint32_t delAckMaxCount,
+                 const TypeId& congControl, const std::string &desc);
+
+protected:
+  virtual void CWndTrace (uint32_t oldValue, uint32_t newValue);
+  virtual void RttTrace (Time oldTime, Time newTime);
+  virtual void BytesInFlightTrace (uint32_t oldValue, uint32_t newValue);
+  virtual void Tx (const Ptr<const Packet> p, const TcpHeader &h, SocketWho who);
+  virtual void Rx (const Ptr<const Packet> p, const TcpHeader &h, SocketWho who);
+  virtual void QueueDrop (SocketWho who);
+  virtual void PhyDrop (SocketWho who);
+  virtual void NormalClose (SocketWho who);
+
+  /**
+   * \brief Update the expected interval at which next packet will be sent
+   */
+  virtual void UpdateExpectedInterval (void);
+
+  virtual void ConfigureEnvironment ();
+  virtual void ConfigureProperties ();
+
+private:
+  uint32_t m_segmentSize;             //!< Segment size
+  uint32_t m_packetSize;              //!< Size of the packets
+  uint32_t m_packets;                 //!< Number of packets
+  EventId  m_event;                   //!< Check event
+  bool     m_initial;                 //!< True on first run
+  uint32_t m_initialCwnd;             //!< Initial value of cWnd
+  uint32_t m_curCwnd;                 //!< Current sender cWnd
+  bool     m_isFullCwndSent;          //!< True if all bytes for that cWnd is sent and sender is waiting for an ACK
+  uint32_t m_bytesInFlight;           //!< Current bytes in flight
+  Time     m_prevTxTime;              //!< Time when Tx was previously called
+  uint16_t m_pacingSsRatio;           //!< Pacing factor during Slow Start
+  uint16_t m_pacingCaRatio;           //!< Pacing factor during Congestion Avoidance
+  uint32_t m_ssThresh;                //!< Slow start threshold
+  bool     m_paceInitialWindow;       //!< True if initial window should be paced
+  uint32_t m_delAckMaxCount;          //!< Delayed ack count for receiver
+  bool     m_isConnAboutToEnd;        //!< True when sender receives a FIN/ACK from receiver
+  Time     m_transmissionStartTime;   //!< Time at which sender starts data transmission
+  Time     m_expectedInterval;        //!< Theoretical estimate of the time at which next packet is scheduled for transmission
+  uint32_t m_packetsSent;             //!< Number of packets sent by sender so far
+  Time     m_nextPacketInterval;      //!< Time maintained by Tx () trace about interval at which next packet will be sent
+  Time     m_tracedRtt;               //!< Traced value of RTT, which may be different from the environment RTT in case of delayed ACKs
+};
+
+
+TcpPacingTest::TcpPacingTest (uint32_t segmentSize,
+                              uint32_t packetSize,
+                              uint32_t packets,
+                              uint16_t pacingSsRatio,
+                              uint16_t pacingCaRatio,
+                              uint32_t ssThresh,
+                              bool paceInitialWindow,
+                              uint32_t delAckMaxCount,
+                              const TypeId &typeId,
+                              const std::string &desc)
+  : TcpGeneralTest (desc),
+    m_segmentSize (segmentSize),
+    m_packetSize (packetSize),
+    m_packets (packets),
+    m_initial (true),
+    m_initialCwnd (10),
+    m_curCwnd (0),
+    m_isFullCwndSent (true),
+    m_bytesInFlight (0),
+    m_prevTxTime (0),
+    m_pacingSsRatio (pacingSsRatio),
+    m_pacingCaRatio (pacingCaRatio),
+    m_ssThresh (ssThresh),
+    m_paceInitialWindow (paceInitialWindow),
+    m_delAckMaxCount (delAckMaxCount),
+    m_isConnAboutToEnd (false),
+    m_transmissionStartTime (Seconds (0)),
+    m_expectedInterval (Seconds (0)),
+    m_packetsSent (0),
+    m_nextPacketInterval (Seconds (0)),
+    m_tracedRtt (Seconds (0))
+{
+  m_congControlTypeId = typeId;
+}
+
+void
+TcpPacingTest::ConfigureEnvironment ()
+{
+  TcpGeneralTest::ConfigureEnvironment ();
+  SetAppPktSize (m_packetSize);
+  SetAppPktCount (m_packets);
+  SetAppPktInterval (NanoSeconds (10));
+  SetMTU (1500);
+  SetTransmitStart (Seconds (0));
+  SetPropagationDelay (MilliSeconds (50));
+}
+
+void TcpPacingTest::ConfigureProperties ()
+{
+  TcpGeneralTest::ConfigureProperties ();
+  SetSegmentSize (SENDER, m_segmentSize);
+  SetInitialSsThresh (SENDER, m_ssThresh);
+  SetInitialCwnd (SENDER, m_initialCwnd);
+  SetPacingStatus (SENDER, true);
+  SetPaceInitialWindow (SENDER, m_paceInitialWindow);
+  SetDelAckMaxCount (RECEIVER, m_delAckMaxCount);
+  NS_LOG_DEBUG ("segSize: " << m_segmentSize << " ssthresh: " << m_ssThresh <<
+                " paceInitialWindow: " << m_paceInitialWindow << " delAckMaxCount " << m_delAckMaxCount);
+}
+
+void
+TcpPacingTest::RttTrace (Time oldTime, Time newTime)
+{
+  NS_LOG_FUNCTION (this << oldTime << newTime);
+  m_tracedRtt = newTime;
+  UpdateExpectedInterval ();
+}
+
+void
+TcpPacingTest::CWndTrace (uint32_t oldValue, uint32_t newValue)
+{
+  NS_LOG_FUNCTION (this << oldValue << newValue);
+  m_curCwnd = newValue;
+  if (m_initial)
+    {
+      m_initial = false;
+    }
+  // CWndTrace () is called after Rx ()
+  // Therefore, call UpdateExpectedInterval () here instead of in Rx ()
+  UpdateExpectedInterval ();
+}
+
+void
+TcpPacingTest::BytesInFlightTrace (uint32_t oldValue, uint32_t newValue)
+{
+  m_bytesInFlight = newValue;
+}
+
+void
+TcpPacingTest::UpdateExpectedInterval (void)
+{
+  double_t factor;
+  Time rtt = 2 * GetPropagationDelay ();
+  if (m_curCwnd < m_ssThresh / 2)
+    {
+      factor = static_cast<double> (m_pacingSsRatio) / 100;
+    }
+  else
+    {
+      factor = static_cast<double> (m_pacingCaRatio) / 100;
+    }
+
+  if (!m_paceInitialWindow && (m_curCwnd == m_initialCwnd * m_segmentSize))
+    {
+      // If initial cwnd is not paced, we expect packet pacing interval to be zero
+      m_expectedInterval = Seconds (0);
+    }
+  else
+    {
+      // Use the estimate according to update equation
+      m_expectedInterval = Seconds ((m_segmentSize * m_tracedRtt.GetSeconds ()) / (factor * m_curCwnd));
+    }
+}
+
+void
+TcpPacingTest::Rx (const Ptr<const Packet> p, const TcpHeader &h, SocketWho who)
+{
+  if (who == SENDER)
+    {
+      uint8_t flags = h.GetFlags ();
+      uint8_t hasFin = flags & TcpHeader::FIN;
+      uint8_t hasAck = flags & TcpHeader::ACK;
+      if (hasFin && hasAck)
+        {
+          m_isConnAboutToEnd = true;
+          NS_LOG_DEBUG ("Sender received a FIN/ACK packet");
+        }
+      else
+        {
+          m_isConnAboutToEnd = false;
+          NS_LOG_DEBUG ("Sender received an ACK packet");
+        }
+    }
+}
+
+void
+TcpPacingTest::Tx (const Ptr<const Packet> p, const TcpHeader &h, SocketWho who)
+{
+  NS_LOG_FUNCTION (this << p << h << who);
+
+  if (who == SENDER)
+    {
+      m_packetsSent++;
+      // Start pacing checks from the second data packet onwards because
+      // an interval to check does not exist for the first data packet.
+      // The first two (non-data) packets correspond to SYN and an
+      // empty ACK, respectively, so start checking after three packets are sent
+      bool beyondInitialDataSegment = (m_packetsSent > 3);
+      Time actualInterval = Simulator::Now () - m_prevTxTime;
+      NS_LOG_DEBUG ("TX sent: packetsSent: " << m_packetsSent << " fullCwnd: " << m_isFullCwndSent << " nearEnd: " <<
+                    m_isConnAboutToEnd << " beyondInitialDataSegment " << beyondInitialDataSegment);
+      if (!m_isFullCwndSent && !m_isConnAboutToEnd && beyondInitialDataSegment)
+        {
+          // Consider a small error margin, and ensure that the actual and expected intervals lie within this error
+          Time errorMargin = NanoSeconds (10);
+          NS_TEST_ASSERT_MSG_LT_OR_EQ (std::abs ((actualInterval - m_nextPacketInterval).GetSeconds ()), errorMargin.GetSeconds (), "Packet delivery in slow start didn't match pacing rate");
+          NS_LOG_DEBUG ("Pacing Check: interval (s): " << actualInterval.GetSeconds () <<
+                        " expected interval (s): " << m_nextPacketInterval.GetSeconds () <<
+                        " difference (s): " << std::abs ((actualInterval - m_nextPacketInterval).GetSeconds ()) <<
+                        " errorMargin (s): " << errorMargin.GetSeconds ());
+        }
+
+      m_prevTxTime = Simulator::Now ();
+      // bytesInFlight isn't updated yet. Its trace is called after Tx
+      // so add an additional m_segmentSize to bytesInFlight
+      uint32_t soonBytesInFlight = m_bytesInFlight + m_segmentSize;
+      bool canPacketBeSent = ((m_curCwnd - soonBytesInFlight) >= m_segmentSize);
+      if (!canPacketBeSent || (m_curCwnd == 0))
+        {
+          m_isFullCwndSent = true;
+        }
+      else
+        {
+          m_isFullCwndSent = false;
+        }
+      m_nextPacketInterval = m_expectedInterval;
+      NS_LOG_DEBUG ("Next expected interval (s): " << m_nextPacketInterval.GetSeconds ());
+    }
+}
+
+void
+TcpPacingTest::QueueDrop (SocketWho who)
+{
+  NS_FATAL_ERROR ("Drop on the queue; cannot validate congestion avoidance");
+}
+
+void
+TcpPacingTest::PhyDrop (SocketWho who)
+{
+  NS_FATAL_ERROR ("Drop on the phy: cannot validate congestion avoidance");
+}
+
+void
+TcpPacingTest::NormalClose (SocketWho who)
+{
+  if (who == SENDER)
+    {
+      m_event.Cancel ();
+    }
+}
+
+/**
+ * \ingroup internet-test
+ * \ingroup tests
+ *
+ * \brief TestSuite for the behavior of TCP pacing
+ */
+class TcpPacingTestSuite : public TestSuite
+{
+public:
+  TcpPacingTestSuite () : TestSuite ("tcp-pacing-test", UNIT)
+  {
+    uint16_t pacingSsRatio = 200;
+    uint16_t pacingCaRatio = 120;
+    uint32_t segmentSize = 1000;
+    uint32_t packetSize = 1000;
+    uint32_t numPackets = 40;
+    uint32_t delAckMaxCount = 1;
+    TypeId tid = TcpNewReno::GetTypeId ();
+    uint32_t ssThresh = 1e9; // default large value
+    bool paceInitialWindow = false;
+    std::string description;
+
+    description = std::string ("Pacing case 1: Slow start only, no initial pacing");
+    AddTestCase (new TcpPacingTest (segmentSize, packetSize, numPackets, pacingSsRatio, pacingCaRatio, ssThresh, paceInitialWindow, delAckMaxCount, tid, description), TestCase::QUICK);
+
+    paceInitialWindow = true;
+    description = std::string ("Pacing case 2: Slow start only, initial pacing");
+    AddTestCase (new TcpPacingTest (segmentSize, packetSize, numPackets, pacingSsRatio, pacingCaRatio, ssThresh, paceInitialWindow, delAckMaxCount, tid, description), TestCase::QUICK);
+
+    // set ssThresh to some smaller value to check that pacing
+    // slows down in second half of slow start, then transitions to CA
+    description = std::string ("Pacing case 3: Slow start, followed by transition to Congestion avoidance, no initial pacing");
+    paceInitialWindow = false;
+    ssThresh = 40;
+    numPackets = 60;
+    AddTestCase (new TcpPacingTest (segmentSize, packetSize, numPackets, pacingSsRatio, pacingCaRatio, ssThresh, paceInitialWindow, delAckMaxCount, tid, description), TestCase::QUICK);
+
+    // Repeat tests, but with more typical delAckMaxCount == 2
+    delAckMaxCount = 2;
+    paceInitialWindow = false;
+    ssThresh = 1e9;
+    numPackets = 40;
+    description = std::string ("Pacing case 4: Slow start only, no initial pacing, delayed ACKs");
+    AddTestCase (new TcpPacingTest (segmentSize, packetSize, numPackets, pacingSsRatio, pacingCaRatio, ssThresh, paceInitialWindow, delAckMaxCount, tid, description), TestCase::QUICK);
+
+    paceInitialWindow = true;
+    description = std::string ("Pacing case 5: Slow start only, initial pacing, delayed ACKs");
+    AddTestCase (new TcpPacingTest (segmentSize, packetSize, numPackets, pacingSsRatio, pacingCaRatio, ssThresh, paceInitialWindow, delAckMaxCount, tid, description), TestCase::QUICK);
+
+    description = std::string ("Pacing case 6: Slow start, followed by transition to Congestion avoidance, no initial pacing, delayed ACKs");
+    paceInitialWindow = false;
+    ssThresh = 40;
+    numPackets = 60;
+    AddTestCase (new TcpPacingTest (segmentSize, packetSize, numPackets, pacingSsRatio, pacingCaRatio, ssThresh, paceInitialWindow, delAckMaxCount, tid, description), TestCase::QUICK);
+  }
+};
+
+static TcpPacingTestSuite g_tcpPacingTest; //!< Static variable for test initialization
diff --git a/src/internet/wscript b/src/internet/wscript
index 38be37089..8bd01e3a7 100644
--- a/src/internet/wscript
+++ b/src/internet/wscript
@@ -305,6 +305,7 @@ def build(bld):
         'test/ipv4-deduplication-test.cc',
         'test/tcp-dctcp-test.cc',
         'test/tcp-syn-connection-failed-test.cc',
+        'test/tcp-pacing-test.cc',
         ]
     # Tests encapsulating example programs should be listed here
     if (bld.env['ENABLE_EXAMPLES']):