Is two-step 802.1AS sensitive to jitter in reality?

[paintspot87]

So I noticed my simulations with clock synchronization added are getting synchronization spikes even though 802.1AS is supposed to guarantee synchronization errors < 1 us. If one message is sent between the sync and followUp message the synchronization goes haywire. Is this behavior expected due to how two-step mode works? It seems very fragile to me. Of course it can simply be fixed by using TAS on the gPTP-messages but what if TAS isn't used to guarantee a jitter-free environment?

I modified showcases/tsn/timesynchronization/gptp OneMasterClock to show this behavior.

[General]
sim-time-limit = 10s
description = "abstract"

# enable time synchronization in all network nodes
*.*.hasTimeSynchronization = true

# all oscillators have a random constant drift
**.tsnClock*.clock.oscillator.typename = "RandomDriftOscillator"
**.oscillator.changeInterval = 12.5ms
**.oscillator.driftRate = uniform(-100ppm, 100ppm)
**.oscillator.driftRateChange = uniform(-1ppm, 1ppm)
**.oscillator.driftRateChangeUpperLimit = 100ppm
**.oscillator.driftRateChangeLowerLimit = -100ppm

# all Ethernet interfaces have 100 Mbps speed
*.*.eth[*].bitrate = 100Mbps

*.visualizer.typename = "IntegratedMultiCanvasVisualizer"
*.visualizer.infoVisualizer[*].displayInfos = true

[Config OneMasterClock]
network = OneMasterClockGptpShowcase
description = "Basic tree topology with one master clock"
*.tsnClock.clock.typename = "IdealClock"
*.*.clock.typename = "SettableClock"

*.*.gptp.syncInitialOffset = 5ms
*.*.gptp.syncInterval = 10ms
*.*.gptp.pdelayInitialOffset = 0s
*.*.gptp.pdelayInterval = 1s
*.tsnDevice1.numApps = 1
*.tsnDevice1.app[0].typename = "UdpSourceApp"
*.tsnDevice1.app[0].source.packetLength = 134B #160B total
*.tsnDevice1.app[0].source.initialProductionOffset = 5002us
*.tsnDevice1.app[0].source.productionInterval = 50ms
*.tsnDevice1.app[0].io.destPort = 1222
*.tsnDevice1.app[0].io.destAddress = "tsnDevice2"

*.tsnDevice2.numApps = 1
*.tsnDevice2.app[0].typename = "UdpSinkApp"
*.tsnDevice2.app[0].io.localPort = 1222

# Uncomment these lines to test out of band time synchronization using the SimpleClockSynchronizer instead of gPTP
#*.tsnClock.gptp.typename = ""
#*.*.gptp.typename = "SimpleClockSynchronizer"
#*.*.gptp.synchronizationInterval = 0.125s
#*.*.gptp.masterClockModule = "^.^.tsnClock.clock"
#*.*.gptp.slaveClockModule = "^.clock"

# TSN clock gPTP master ports
*.tsnClock.gptp.masterPorts = ["eth0"]

# TSN switch gPTP bridge master ports
*.tsnSwitch.gptp.masterPorts = ["eth1", "eth2"]

# Set all reference clocks to master clock so the time difference can be visualized
**.referenceClock = "tsnClock.clock"

# data link visualizer displays gPTP time synchronization packets
*.visualizer.dataLinkVisualizer[0].displayLinks = true
*.visualizer.dataLinkVisualizer[0].activityLevel = "protocol"
*.visualizer.dataLinkVisualizer[0].packetFilter = "GptpSync"
*.visualizer.dataLinkVisualizer[0].lineColor = "blue2"

*.visualizer.numInfoVisualizers = 3
*.visualizer.infoVisualizer[0].modules = "*.tsnClock.clock"
*.tsnClock.clock.displayStringTextFormat = "time: %T"
*.visualizer.infoVisualizer[1].modules = "*.tsnSwitch.clock"
*.visualizer.infoVisualizer[1].placementHint = "topLeft"
*.visualizer.infoVisualizer[2].modules = "*.tsnDevice*.clock"
*.visualizer.infoVisualizer[2].placementHint = "bottom"
*.tsnDevice*.clock.displayStringTextFormat = "diff: %d"
*.tsnSwitch.clock.displayStringTextFormat = "diff: %d"

Result with packet sent between sync and follow-up:

Result when commenting out #*.tsnDevice1.app[0].source.initialProductionOffset = 5002us

[ZoltanBojthe]

This is an inet issue, currently fixed in inet at topic/bz/802.1as branch.

[paintspot87]

Perfect, thank you so much. I merged the parts and now there are no synchronization spike issues anymore.