an innovative traffic management scheme for deterministic
TRANSCRIPT
An innovative traffic management scheme for deterministic/event-
based communications in automotive applications with a focus on
Automated Driving Applications
Giancarlo Vasta, Magneti Marelli, [email protected]
Lucia Lo Bello, University of Catania, [email protected]
October 10, 2018
2018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day
Presentation outline
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day
■ Event-driven traffic in automated vehicles
■ Support offered to event-driven traffic in TSN standards
■ A novel approach to deal with event-driven traffic
■ Performance evaluation
■ Automated Driving application
■ Summary and conclusions
2
The need for Event-driven Traffic Support
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day
Fact 1: The external world in which the cars move is non-deterministic
Unpredictable events may occur and determine unforeseen situations.
Fact 2: Autonomous vehicles need to properly react to such situations
The communication system has to support event-driven transmissions with
minimum delay.
Question 1: How to handle event-driven
traffic with real-time constraints?
Question 2: How and under which
conditions is it possible to guarantee
a latency bound for event-driven flows?
Let’s have a look on the TSN standards…
3
TSN support for Event-driven Traffic
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day
Fact 3: TSN standards offer a limited support for event-driven real-time
traffic
� Some shaping mechanisms have been defined, but
� they do not care much about the latency of event-driven traffic
or
� they can be complex
� The easiest way to deal with event-driven traffic is to serve it in the Best-effort
class, but
� no guarantees, even in the case the event-driven traffic has a known
minimum inter-arrival time, can be provided.
4
This presentation is about
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day
An innovative traffic management scheme for IEEE 802.1Q bridges and end
nodes that
� introduces explicit support for the Event-Driven (ED) real-time traffic
� enables the transmission of ED traffic over TSN networks while
● maintaining the support for Scheduled Traffic
● providing delay guarantees to the Stream Reservation classes.
The proposed approach:
� works at the MAC layer
� relies on the forwarding mechanisms defined in the IEEE 802.1Q-2018
standard, with suitable modifications
5
System model
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day
Traffic type Characteristics
Scheduled traffic
High priority real-time traffic
transmitted according to a
time schedule (time-driven)
with no interference from
other traffic.
Stream Reservation Periodic, guaranteed
Event-driven traffic
Aperiodic bursts, generated
by events, with real-time
constraints
Best-effort traffic No guarantees,
performance is statistical
6
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day
Main features
� Scheduled Traffic is handled through the gate mechanism defined in the IEEE
802.1Qbv
� ED bursts are handled with a novel traffic management
Low latency for ED flows is achieved
� If the minimum interarrival time for ED bursts and the maximum burst size are
known, the maximum E2E latency for ED traffic is bounded and can be
calculated.
� E2E latency for SR Class A and Class B are bounded and guaranteed.
� Feasibility: The proposed approach can be implemented on devices that
support TSN standards. No special hardware modifications to the standard
devices are required.
Advantages of the proposed solution
7
Simulations
Simulations were run using the OMNeT++ and the INET framework, modified
so as to model the TSN protocols (i.e., IEEE 802.1Q-2018, etc.)
Aims
• Evaluate the E2E performance of the proposed approach in realistic
automotive scenarios
• E2E Latency performance with different data rates (i.e., 100Mbps, 1Gbps,
10Gbps).
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 8
The proposed solution was compared with an alternative approach, that
handles the ED traffic as best-effort, in the highest priority queue for that class.
ED TrafficED Traffic
Alternative approach
ST
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 9
Simulations
TSN Switch
ECU
• Each node transmits sensor data to the ECU
• 3 configurations:
• 100Mbps: 30 fps compressed video streams,
relative deadline 33ms.
• 1Gbps: 60 fps compressed video streams,
relative deadline 10ms .
• 10Gbps: 30 fps uncompressed video streams,
relative deadline 10ms .
Node
- Radars
- Lidars
- Ultrasonics
- Cameras
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 10
Simulation scenarios
Flow Number of flows Total MAC Bandwidth
Lidar 4 0.93 Mbps
Ultrasonic 4 0.23 Mbps
ADAS Sensors 4 Max 34 Mbps
30 fps compressed Video 4 ~52 Mbps
Case A: 100 Mbps Scenario – ADAS
Flow Number of flows Total MAC Bandwidth
Lidar 4 0.93 Mbps
Ultrasonic 4 0.23 Mbps
ADAS Sensors 4 Max 34 Mbps
60 fps compressed Video 4 ~103 Mbps
Case B: 1 Gbps Scenario – ADAS
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 11
Scenarios
Flow Number of flows Total MAC Bandwidth
Lidar 4 0.93 Mbps
Ultrasonic 4 0.23 Mbps
ADAS Sensors 4 Max 34 Mbps
30 fps uncompressed Video 4 ~7 Gbps
Case C: 10 Gbps Scenario – Automated Driving
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 12
Scenarios
Flow Parameters and Mapping
Flow Length Rel.
Deadline
Sampling
Time
Arrival
pattern
Traffic Class
Lidar 250 B 10 ms 10 ms Periodic ST
Ultrasonic 100 B 20 ms 20 ms Periodic ST
ADAS Sensors ~10 KB 10 ms [10-100] ms Event-driven ED/BE
30fps Video - ADAS
(100 Mbps)~43 KB 33 ms 33 ms Periodic SR_A
60fps Video - ADAS
(1 Gbps)~43 KB 10 ms 16 ms Periodic SR_A
Raw 30fps Video –
Automated Driving
(10 Gbps)
6.9 MB 33 ms 33 ms Periodic SR_A
Class A video frames are segmented in small Ethernet frames, each one
transmitted by the application every 125us.
Performance metrics: End-to-end latency, defined as:
reception time – generation time (measured at the application)
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 13
Event-driven flows
The proposed approach
- halves the maximum E2E latency
- reduces the average message latency by 870us
5119
19742332
1104
0
1000
2000
3000
4000
5000
6000
Max E2E-Latency Avg E2E-Latency
us
ED Flows: Message latency
Alternative Approach Proposed Approach
SR 52%
ED 33%
ST 2%
Unused 13%
Case A: Workload vs. bandwidth
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 14
Simulation Results – Case A: 100 Mbps scenario
Event-driven flows – End-to-end latency distribution
Alternative Approach
Proposed Approach
The ED traffic with the
proposed approach obtains
• lower latency values
• lower latency variability
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 15
Simulation Results – Case A: 100 Mbps scenario
Simulation Results – Case B: 1 Gbps scenario
Event-driven flows
The max latency of event-driven messages is almost the same in both configurations.
Reason:
Very low workload, the SR workload is 11% of the available bandwidth.
SR 11%ED 3%
Other 0,20%
Unused 86%
Case B: Workload vs. bandwidth
218
111
192
105
0
50
100
150
200
250
Max E2E-Latency Avg E2E-Latency
us
ED Flows: Message latency
Alternative Approach Proposed Approach
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 16
Simulation Results – Case B: 1 Gbps scenario
Event-driven flows – End-to-end latency distribution
Proposed Approach
Alternative ApproachThe ED traffic with the
proposed approach obtains
• lower latency values
• lower latency variability
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 17
Simulation Results – Case C: 10 Gbps scenario
Event-driven flows
Under high SR workload (69% in this scenario)
the proposed approach significantly reduces the
latency of event-driven traffic.
SR 69%
ED 0,33%
Other 0,02%
Unused 31%
Case C: Workload vs. bandwidth
63
35
20
15
0
10
20
30
40
50
60
70
Max E2E-Latency Avg E2E-Latency
us
ED Flows: Message latency
Alternative Approach Proposed Approach
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 18
Simulation Results – Case C: 10 Gbps scenario
Event-driven flows – End-to-end latency distribution
Proposed Approach
Alternative Approach
The ED traffic with the
proposed approach obtains:
• lower latency values
• lower latency variability
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 19
Other traffic classes – Case A: 100 Mbps scenario
31332 32466
0
5000
10000
15000
20000
25000
30000
35000
AppMessage Max
us
Video Flows: Max latencyper video frame
Alternative Approach Proposed Approach
• Scheduled traffic is unaltered.
• SR Class A traffic (i.e., video streams) meets the relative deadline of 33ms
per video frame with both approaches.
Rel.
Deadline
33ms50
26
50
26
0
10
20
30
40
50
60
70
80
90
100
Lidar Ultrasonic
us
ST Flows
Alternative Approach Proposed Approach
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 20
Other traffic classes – Case A: 100 Mbps scenario
Video flows (SR_A) – End-to-end latency distribution
Alternative Approach
Proposed Approach
• The proposed approach obtains a
longer tail, but the difference between
the maximum values is around 3%
and affects only 0.3% of the video
frames.
• The relative deadline (33ms) is met in
both cases.
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 21
Other traffic classes – Case B: 1 Gbps scenario
• Scheduled traffic is unaltered.
• SR Class A traffic (i.e., video streams) meets the relative deadline of 10ms
per video frame in both approaches.
7993 8132
0
2000
4000
6000
8000
10000
AppMessage Max
us
Video Flows: Max latencyper video frame
Alternative Approach Proposed Approach
Rel.
Deadline 10
ms9,68
7,28
9,68
7,28
0
2
4
6
8
10
12
14
16
18
20
Lidar Ultrasonic
us
ST Flows
Alternative Approach Proposed Approach
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 22
Other traffic classes – Case B: 1 Gbps scenario
Video flows (SR_A) – End-to-end latency distribution
Proposed Approach
Alternative Approach
The proposed approach
obtains slightly higher latency
values and variability for SR
flows, but the difference is
minor.
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 23
Other traffic classes – Case C: 10 Gbps scenario
• Scheduled traffic is unaltered.
• SR Class A traffic (i.e., video streams) meets the relative deadline of 33ms
per video frame in both approaches.
31005 31005
0
5000
10000
15000
20000
25000
30000
35000
AppMessage Maxu
s
Video Flows: Max latencyper video frame
Alternative Approach Proposed Approach
Rel.
Deadline 33
ms5,65 5,415,65 5,41
0
1
2
3
4
5
6
7
8
9
10
Lidar Ultrasonic
us
ST Flows
Alternative Approach Proposed Approach
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 24
Other traffic classes – Case C: 10 Gbps scenario
Video Flows (SR_A) – End-to-end latency distribution
Proposed Approach
Alternative Approach
Plenty of bandwidth, so no
difference between the two
approaches.
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day 25
October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day
� The proposed approach provides event-driven traffic with
� very low maximum latency values
� very low latency variability.
� In addition, the proposed approach
● does not affect Scheduled traffic
● has a very limited impact on the E2E maximum latency for Stream
Reservation traffic
Summarizing
26
Application on Autonomous Driving: Smart Corner™
27October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day
Automated Driving architecture
28October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day
Lidar
Radar
Camera
UltraSound
Component
/ HW Layer
Function / SW Layer
V2X Contr. Unit
Automated Driving
Domain Controller
Se
nso
rs
Sensor specific
ECU (?) *
PWT ECU
Steering ECU
Brakes ECU
Suspension ECU
GNSS Receiver
Environment
Data Sensing /
Receiving
Environment
Data Sensing /
Receiving
Object
Detection
Object
Detection
Sensor
Fusion
Sensor
Fusion
MappingMapping
Data Fusion /
Environment
Model
Data Fusion /
Environment
Model
DecisionDecision
Motion & Actuation
Planning & Control
Motion & Actuation
Planning & Control
ActuationActuation
Understanding the
context around
the vehicle
Planning and
executing
motion
Perception Think & Decision Actuation
(Device specific
driver and SW)
Int and Ext.
V2X Led Lights
* Sensor specific ECUs used instead of Automated Driving Domain Controller for Individual ADAS functions
Magneti Marelli Autonomous Vehicles
29October 10, 20182018 IEEE Standards Association Ethernet & IP @ Automotive Technology Day
Autonomous Vehicle for
Parking scenario (level 4)
Autonomous Vehicle for
Highway scenario (level 3)
Autonomous Vehicle for
Urban scenario (level 4)
Autonomous Vehicle