Markus Fiedler: Internet Performance Issues for Safety-Critical Applications
Internet Performance Issues for Safety-Critical Applications
Markus Fiedler
Blekinge Institute of Technology School of Engineering
Dept. of Telecommunication Systems
Internet Issues (1)
• Its very nature
• A network of networks
• Connectivity – the only concern?
• “Can you ping?” (If yes, everything is OK.)
• Shared resources and the best-effort paradigm
– “Jantelagen”: “Du skall icke tro att du är bättre än vi.”
– No guarantees
• IP (layer 3) uses datagrams
– No explicit quality feedback between the layers
• TCP (layer 4) “times out”
Markus Fiedler: Internet Performance Issues for Safety-Critical Applications
Internet Issues (2)
• Few QoS standardization and handling efforts – IntServ does not scale
– Island-type solutions
• MPLS (DiffServ) in the core
• Industrial real-time Ethernet at the edge
• QoS handling – if any – is mainly left to the end systems/application
– RTCP (layer 7): monitors quality for real-time applications – TCP (layer 4): loss and sequence recovery at the cost of
delay
• Motto: Help yourself!
... And Some Consequences
• Packet streams
– Meet braking transport capacity limitations (bottlenecks)
– Interfere with (= brake) each other
• User-perceived Quality of Service (QoS) in terms of – Speed: Too low and varying delay and jitter – Accuracy: Too low losses
– Reliability: Not dependable time-/drop-outs
• Perceived speed < installed capacity
Markus Fiedler: Internet Performance Issues for Safety-Critical Applications
Perceived Speed
Low capacity
High cap.
Server
Switch Client
Ethernet
Transmission time Processing time ()
Queuing time Entity level Transmission
level
Legend:
time
Travel Length/capacity Client
Effective speed
Perceived Speed – Shared (1)
High cap.
Server
Switch Client
Ethernet
Transmission time Processing time ()
Queuing time
Entity level Transmission
level
Legend:
time
Travel Length/capacity Client
Effective speed
Low capacity Collision
Markus Fiedler: Internet Performance Issues for Safety-Critical Applications
Perceived Speed – Shared (2)
Low capacity
High cap.
Server
Switch Client
Transmission time Processing time ()
Queuing time Entity level Transmission
level
Legend:
time
Travel Length/capacity Client
Effective speed
Perceived Speed – Shaped
Low capacity
High cap.
Server
Switch Client
Transmission time Processing time ()
Queuing time Entity level Transmission
level
Legend:
time
Travel Length/capacity Client
Effective speed
Speed adaptation
Markus Fiedler: Internet Performance Issues for Safety-Critical Applications
Loss Due To Overload
Low capacity
High cap.
Server
Switch Client
Ethernet
Transmission time Processing time ()
Queuing time Entity level Transmission
level
Legend:
time
Travel Length/capacity Client
Dependability
Low capacity
Server
Switch Client
Ethernet
Transmission time Processing time ()
Queuing time Entity level Transmission
level
Legend:
time
Travel Length/capacity Client
Markus Fiedler: Internet Performance Issues for Safety-Critical Applications
Some QoS Degradation Measures (1)
• One-way delay
– Problematic due to time synchronization
• NTP useful in local environments
• GPS might help
• Round-trip delay
– Determines download times
– Approximately available through ping
• Active measurement: implies extra load; just probing – Upper bound for the one-way delay
• Asymmetrical delays are quite normal
Some QoS Degradation Measures (2)
• Delay jitter
– Comparably simple to obtain – Trend analysis required
• Loss
– To be observed on upper layers
• Reliability
– Real-time applications: Frequency of time-outs – General: Relative downtime
Markus Fiedler: Internet Performance Issues for Safety-Critical Applications
Safety-Critical Applications
• Needs:
– Minimal delay and (additional) jitter
• Keep timing relationships as good as possible – No loss
– Information = feedback in case of problems
• Characteristics:
– Streaming: amount ∆L within each ∆T
• Control; voice; video; ...
– Messaging: amount L to be received within T
• Alarms; notifications; ...
– Interactive: amounts LCS + LSCto be received within T
Speed Considerations
• Close to limiting factor: lack of transport capacity
⇒ Queuing ⇒ delay, jitter
⇒ Loss
• Calculation of end-to-end-perceived speed – Time synchronization upon arrival of first packet
• As perceived by the receiver
– Bandwidth changes along the way through the network reflect jitter and loss
– Focus on streaming services
• Provides speed information even for messaging/interactive services
Markus Fiedler: Internet Performance Issues for Safety-Critical Applications
End-To-End Perceived Speed – Ideal Case
∆L
Delay of 1stpacket
∆T
∆L/∆T
End-To-End Perceived Speed – Jitter (1)
Delay of 1stpacket
∆T
∆L ∆L/∆T
Delayed packet speed change
Markus Fiedler: Internet Performance Issues for Safety-Critical Applications
End-To-End Perceived Speed – Jitter (2)
Delay of 1stpacket
∆T
∆L ∆L/∆T
End-To-End Perceived Speed – Loss
Delay of 1stpacket
∆T
∆L L/ ∆T
Lost packet speed change
Markus Fiedler: Internet Performance Issues for Safety-Critical Applications
”Kilroy was here”
• http://www.kilroywashere.org
• End-to-end-perceived speed measurements and histograms tell about
– Type of bottleneck – Severity of disturbance
• ”Kilroy indicator” = histogram@output – histogram@input – Observation window ∆W
– Averaging interval ∆T – Speed discretization ∆R
Shared Bottleneck – Example
In
Speed Out
Time interval
1
Speed
H
1
Speed
H
1
-1
∆H
Speed
Markus Fiedler: Internet Performance Issues for Safety-Critical Applications
Shaping Bottleneck – Example
In
Speed Out
Time interval
1
Speed
H
1
Speed
H
1
-1
∆H
Speed
Video Experience of an Artificial Bottleneck
Switch
Switch Internet
Karlskrona (SE) Generator
Consumer 10 Mbps half-duplex
Würzburg (D)
Distur- bing UDP traffic 010Mbps
Speed histogram
Karlskrona Würzburg ”Kilroy”
Difference
Measure- ment point Würzburg Measure- ment point Karlskrona
Markus Fiedler: Internet Performance Issues for Safety-Critical Applications
A Closer Look on Audio
• One packet à 492 B each 60 ms
• Inter-packet jitter
– Observation window = 1 min
– Jitter = Packet inter-arrival time – 60 ms – Discretization = 1 ms
• Speed
– Observation window = 1 min – Averaging interval = 60 ms – Speed discretization = 10 kbps
• Different levels of disturbance
Jitter – Disturbance
0 Mbps 6 Mbps
-0,15 -0,1 -0,05 0 0,05 0,1 0,15 0,2 0,25 0,3
-0,06 -0,04 -0,02 0 0,02 0,04
Jitter [s]
Rel. frequency
Jitter in Jitter out Difference
-0,15 -0,1 -0,05 0 0,05 0,1 0,15 0,2 0,25 0,3
-0,06 -0,04 -0,02 0 0,02 0,04 0,06
Jitter [s]
Rel. frequency
Jitter in Jitter out Difference
-0,15 -0,1 -0,05 0 0,05 0,1 0,15
-0,06 -0,04 -0,02 0 0,02 0,04
Jitter [s]
Rel. frequency
Difference
-0,15 -0,1 -0,05 0 0,05 0,1 0,15
-0,06 -0,04 -0,02 0 0,02 0,04 0,06
Jitter [s]
Rel. frequency
Difference
Markus Fiedler: Internet Performance Issues for Safety-Critical Applications
Speed – Disturbance 0 Mbps
-20%
0%
20%
40%
60%
80%
100%
0 20000 40000 60000 80000 100000 120000 140000 160000
Sender Receiver Difference Difference
-2,5%
-2,0%
-1,5%
-1,0%
-0,5%
0,0%
0,5%
1,0%
1,5%
0 20000 40000 60000 80000 100000 120000 140000 160000
bps→
Speed – Disturbance 6 Mbps
-20%
0%
20%
40%
60%
80%
100%
0 20000 40000 60000 80000 100000 120000 140000 160000
Sender Receiver Difference
Difference
-12,5%
-10,0%
-7,5%
-5,0%
-2,5%
0,0%
2,5%
5,0%
7,5%
0 20000 40000 60000 80000 100000 120000 140000 160000
bps→
Markus Fiedler: Internet Performance Issues for Safety-Critical Applications
Speed Comparisons
+ Reflects flow of packets
+ Takes packet lengths into account + Comparable to installed capacities + Comparable to expected behaviour
+ See whether packets come ”too late”
+ Comparable to behaviour at inlet
• Requires exchange of speed histograms + Feedback facility
− Dependency on first packet
− Short flows
− Granularity/discretization
Conclusion And Open Issues
• Quality of Service problems in terms of – Speed
– Accuracy – Reliability
are well reflected in speed changes
• Important: Time scale of interest
• Time synchronization issue and one-way delays via Internet
– Measurement – Guarantee
Markus Fiedler: Internet Performance Issues for Safety-Critical Applications
Internet Performance Issues for Safety-Critical Applications
Markus Fiedler Thanks for listening ☺☺☺☺
Any questions?