Realistic simulation scenarios

README.md

@@ -2,8 +2,12 @@
 
 * DST gossipsub test node
 * incl shadow simulation setup
+* incl awk scripts for detailed analysis
 
-## Shadow example
+Three simulation cases are supportes
+
+## Scenario1 - Basic Simulation Scenario 
+### Shadow example
 
 ```sh
 nimble install -dy
@@ -22,3 +26,44 @@ nim -d:release c plotter
 ```
 
 The dependencies will be installed in the `nimbledeps` folder, which enables easy tweaking
+
+
+## Scenario2 - (P2P-Research Branch) - 
+Homogeneous nodes/links (100 Mbps bandwidth and 100ms latency)
+supports variable network/message size, variable number of publishers and message fragments
+automatically updates shadow.yaml file to accomodate different network sizes 
+awk scripts for detailed analysis
+
+```sh
+cd shadow
+#the run.sh script is automated to meet different experiment needs, use ./run.sh <num_runs num_peers msg_size num_fragments>
+#The below example runs the simulation twice for a 1000 node network. each publisher publishes a 15000 bytes messages, and every message is partitioned into 4 fragments. Total 10 publishers use  
+
+./run.sh 2 1000 15000 4 10
+# The number of nodes is maintained in the shadow.yaml file, and automatically updated by run.sh.
+# The output files latencies(x), stats(x) and shadowlog(x) carries the outputs for each simulation run.
+# The summary_dontwant.awk, summary_latency.awk, summary_latency_large.awk, and summary_shadowlog.awk parse the output files.
+# The run.sh script automatically calls these files to display the output
+# a temperary data.shadow folder is created for each simulation and removed by the run.sh after the simulation is over
+```
+
+## Scenario3 - (Realistic-Scenarios Branch) - 
+Heterogeneous nodes/links (bandwidth/latency/packet_loss_ratio controllable through run.sh)
+run.sh uses topogen.py to create a realistic gml file to emulate a real-world network scenario
+supports variable network/message size, variable number of publishers and message fragments
+automatically generates network_topology.gml and shadow.yaml files to accomodate different network sizes 
+awk scripts called from run.sh for detailed analysis
+
+```sh
+cd shadow
+#The following sample command runs simulation 1 time, for a 1000 node network. Each published message size
+#is 15KB (no-fragmentation), peer bandwidth varies between 50-130 Mbps, Latency between 60-160ms, and
+#bandwidth,latency is roughly distributed in five different groups. 
+
+./run.sh 1 1000 15000 1 10 50 130 60 160 5 0.0
+# The number of nodes is maintained in the shadow.yaml file, and automatically generated by run.sh.
+# The output files latencies(x), stats(x) and shadowlog(x) carries the outputs for each simulation run.
+# The summary_dontwant.awk, summary_latency.awk, summary_latency_large.awk, and summary_shadowlog.awk parse the output files.
+# The run.sh script automatically calls these files to display the output
+# a temperary data.shadow folder is created for each simulation and removed by the run.sh after the simulation is over
+```

shadow/main.nim

@@ -1,12 +1,22 @@
 import stew/endians2, stew/byteutils, tables, strutils, os
 import libp2p, libp2p/protocols/pubsub/rpc/messages
 import libp2p/muxers/mplex/lpchannel, libp2p/protocols/ping
+#import libp2p/protocols/pubsub/pubsubpeer
+
 import chronos
 import sequtils, hashes, math, metrics
 from times import getTime, toUnix, fromUnix, `-`, initTime, `$`, inMilliseconds
 from nativesockets import getHostname
 
-const chunks = 1
+#These parameters are passed from yaml file, and each defined peer may receive different parameters (e.g. message size)
+var
+  publisherCount = parseInt(getEnv("PUBLISHERS"))
+  msg_size = parseInt(getEnv("MSG_SIZE")) 
+  chunks = parseInt(getEnv("FRAGMENTS"))
+
+#we experiment with upto 10 fragments. 1 means, the messages are not fragmented
+if chunks < 1 or chunks > 10:     
+  chunks = 1
 
 proc msgIdProvider(m: Message): Result[MessageId, ValidationResult] =
   return ok(($m.data.hash).toBytes())
@@ -15,9 +25,7 @@ proc main {.async.} =
   let
     hostname = getHostname()
     myId = parseInt(hostname[4..^1])
-    #publisherCount = client.param(int, "publisher_count")
-    publisherCount = 10
-    isPublisher = myId <= publisherCount
+    isPublisher = myId <= publisherCount      #need to adjust is publishers ldont start from peer1
     #isAttacker = (not isPublisher) and myId - publisherCount <= client.param(int, "attacker_count")
     isAttacker = false
     rng = libp2p.newRng()
@@ -45,7 +53,7 @@ proc main {.async.} =
       anonymize = true,
       )
     pingProtocol = Ping.new(rng=rng)
-  gossipSub.parameters.floodPublish = false
+  gossipSub.parameters.floodPublish = false 
   #gossipSub.parameters.lazyPushThreshold = 1_000_000_000
   #gossipSub.parameters.lazyPushThreshold = 0
   gossipSub.parameters.opportunisticGraftThreshold = -10000
@@ -79,6 +87,9 @@ proc main {.async.} =
       sentNanosecs = nanoseconds(sentMoment - seconds(sentMoment.seconds))
       sentDate = initTime(sentMoment.seconds, sentNanosecs)
       diff = getTime() - sentDate
+
+    #  pubId = byte(data[11])
+
     echo sentUint, " milliseconds: ", diff.inMilliseconds()
 
 
@@ -121,7 +132,7 @@ proc main {.async.} =
   let connectTo = parseInt(getEnv("CONNECTTO"))
   var connected = 0
   for peerInfo in peersInfo:
-    if connected >= connectTo: break
+    if connected >= connectTo+2: break
     let tAddress = "peer" & $peerInfo & ":5000"
     echo tAddress
     let addrs = resolveTAddress(tAddress).mapIt(MultiAddress.init(it).tryGet())
@@ -137,24 +148,64 @@ proc main {.async.} =
   #  warmupMessages = client.param(int, "warmup_messages")
   #startOfTest = Moment.now() + milliseconds(warmupMessages * maxMessageDelay div 2)
 
-  await sleepAsync(10.seconds)
-  echo "Mesh size: ", gossipSub.mesh.getOrDefault("test").len
+  await sleepAsync(12.seconds)
+  echo "Mesh size: ", gossipSub.mesh.getOrDefault("test").len, 
+      ", Total Peers Known : ", gossipSub.gossipsub.getOrDefault("test").len,
+      ", Direct Peers : ", gossipSub.subscribedDirectPeers.getOrDefault("test").len,
+      ", Fanout", gossipSub.fanout.getOrDefault("test").len, 
+      ", Heartbeat : ", gossipSub.parameters.heartbeatInterval.milliseconds
+
+  await sleepAsync(5.seconds)  
 
-  for msg in 0 ..< 10:#client.param(int, "message_count"):
-    await sleepAsync(12.seconds)
-    if msg mod publisherCount == myId - 1:
-    #if myId == 1:
+  # Actual message publishing, one message published every 3 seconds
+  # First 1-2 messages take longer than expected time due to low cwnd. 
+  # warmup_messages can set cwnd to a desired level. or alternatively, warmup messages can be set to 0
+  let
+    warmup_messages = 2 
+    #shadow.yaml defines peers with changing latency/bandwith. In the current arrangement all the publishers 
+    #will get different latency/bandwidth
+    pubStart = 4                                                       
+    pubEnd = pubStart + publisherCount + warmup_messages
+
+
+  #we send warmup_messages for adjusting TCP cwnd  
+  for i in pubStart..<(pubStart + warmup_messages):
+    await sleepAsync(2.seconds)
+    if i == myId:
+      #two warmup messages for cwnd raising
+      let
+          now = getTime()
+          nowInt = seconds(now.toUnix()) + nanoseconds(times.nanosecond(now))
+      var nowBytes = @(toBytesLE(uint64(nowInt.nanoseconds))) & newSeq[byte](msg_size)
+      doAssert((await gossipSub.publish("test", nowBytes)) > 0)
+  #done sending warmup_messages , wait for short time
+  await sleepAsync(5.seconds)
+
+
+  #We now send publisher_count messages
+  for msg in (pubStart + warmup_messages) .. pubEnd:#client.param(int, "message_count"):
+    await sleepAsync(3.seconds)
+    if msg mod (pubEnd+1) == myId:
       let
         now = getTime()
         nowInt = seconds(now.toUnix()) + nanoseconds(times.nanosecond(now))
-      #var nowBytes = @(toBytesLE(uint64(nowInt.nanoseconds))) & newSeq[byte](500_000 div chunks)
-      var nowBytes = @(toBytesLE(uint64(nowInt.nanoseconds))) & newSeq[byte](50)
-      #echo "sending ", uint64(nowInt.nanoseconds)
+    #[    
+      if chunks == 1:
+        var nowBytes = @(toBytesLE(uint64(nowInt.nanoseconds))) & newSeq[byte](50000)
+      else:
+        var nowBytes = @(toBytesLE(uint64(nowInt.nanoseconds))) & newSeq[byte](500_000 div chunks)
+    ]#
+      var nowBytes = @(toBytesLE(uint64(nowInt.nanoseconds))) & newSeq[byte](msg_size div chunks)
       for chunk in 0..<chunks:
         nowBytes[10] = byte(chunk)
         doAssert((await gossipSub.publish("test", nowBytes)) > 0)
-
-  #echo "BW: ", libp2p_protocols_bytes.value(labelValues=["/meshsub/1.1.0", "in"]) + libp2p_protocols_bytes.value(labelValues=["/meshsub/1.1.0", "out"])
-  #echo "DUPS: ", libp2p_gossipsub_duplicate.value(), " / ", libp2p_gossipsub_received.value()
-
+      echo "Done Publishing ", nowInt.nanoseconds
+
+  #we need to export these counters from gossipsub.nim
+  echo "statcounters: dup_during_validation ", libp2p_gossipsub_duplicate_during_validation.value(),
+       "\tidontwant_saves ", libp2p_gossipsub_idontwant_saved_messages.value(),
+       "\tdup_received ", libp2p_gossipsub_duplicate.value(),
+       "\tUnique_msg_received ", libp2p_gossipsub_received.value(),
+       "\tStaggered_Saves ", libp2p_gossipsub_staggerDontWantSave.value(),
+       "\tDontWant_IN_Stagger ", libp2p_gossipsub_staggerDontWantSave2.value()
 waitFor(main())

shadow/run.sh

@@ -1,8 +1,65 @@
 #!/bin/sh
-
 set -e
 
-nim c -d:chronicles_colors=None --threads:on -d:metrics -d:libp2p_network_protocols_metrics -d:release main
-rm -rf shadow.data/
-shadow shadow.yaml
-grep -rne 'milliseconds\|BW' shadow.data/ > latencies
+if [ $# -ne 11 ]; then
+    echo "Usage: $0 <runs> <nodes> <Message_size> <num_fragment> <num_publishers>
+            <min_bandwidth> <max_bandwidth> <min_latency> <max_latency> <anchor_stages <packet_loss>>"
+
+    echo "The following sample command runs simulation 1 time, for a 1000 node network. Each published message size \
+            is 15KB (no-fragmentation), peer bandwidth varies between 50-130 Mbps, Latency between 60-160ms, and \
+            bandwidth,latency is roughly distributed in five different groups. \
+            see the generated network_topology.gml and shadow.yaml for peers/edges details"   
+
+    echo "$0 1 1000 15000 1 10 50 130 60 160 5 0.0" 
+    exit 1
+fi
+
+runs="$1"			        #number of simulation runs
+nodes="$2"			        #number of nodes to simulate
+msg_size="$3"			    #message size to use (in bytes)
+num_frag="$4"			    #number of fragments per message (1 for no fragmentation)
+num_publishers="$5"         #number of publishers 
+min_bandwidth="$6"
+max_bandwidth="$7"
+min_latency="$8"
+max_latency="$9"
+steps="${10}"
+pkt_loss="${11}"
+
+connect_to=5                #number of peers we connect with to form full message mesh
+
+
+#topogen.py uses networkx module from python to generate gml and yaml files
+PYTHON=$(which python3 || which python)
+
+if [ -z "$PYTHON" ]; then
+    echo "Error: Python, Networkx is required for topology files generation."
+    exit 1
+fi
+
+"$PYTHON" topogen.py $nodes $min_bandwidth $max_bandwidth $min_latency $max_latency $steps $pkt_loss $msg_size $num_frag $num_publishers
+
+
+
+rm -f shadowlog* latencies* stats* main && rm -rf shadow.data/
+nim c -d:chronicles_colors=None --threads:on -d:metrics -d:libp2p_network_protocols_metrics -d:release main 
+
+for i in $(seq $runs); do
+    echo "Running for turn "$i
+    shadow shadow.yaml > shadowlog$i && 
+        grep -rne 'milliseconds\|BW' shadow.data/ > latencies$i && 
+        grep -rne 'statcounters:' shadow.data/ > stats$i
+    #uncomment to to receive every nodes log in shadow data (only if runs == 1, or change data directory in yaml file)
+    #rm -rf shadow.data/
+done
+
+for i in $(seq $runs); do
+    echo "Summary for turn "$i
+    if [ "$msg_size" -lt 1000 ]; then
+    	awk -f summary_latency.awk latencies$i		#precise per hop coverage for short messages only
+    else
+	awk -f summary_latency_large.awk latencies$i	#estimated coverage for large messages (TxTime adds to latency)
+    fi
+    awk -f summary_shadowlog.awk shadowlog$i
+    awk -f summary_dontwant.awk stats$i    
+done