Evaluation of Online Schedule Synthesis Algorithms for Time-based - - PowerPoint PPT Presentation

▶
evaluation of online schedule synthesis algorithms for
SMART_READER_LITE
LIVE PREVIEW

Evaluation of Online Schedule Synthesis Algorithms for Time-based - - PowerPoint PPT Presentation

Dec 14, 2022 48 likes •302 views

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Evaluation of Online Schedule Synthesis Algorithms for Time-based Scheduled Time Sensitive Networks Final talk for the Masters Thesis by

slide-1
SLIDE 1

Chair of Network Architectures and Services Department of Informatics Technical University of Munich

Evaluation of Online Schedule Synthesis Algorithms for Time-based Scheduled Time Sensitive Networks

Final talk for the Master’s Thesis by

Alexander Mildner

advised by Max Helm, Benedikt Jaeger, Dr. Marcel Wagner (Intel), Hector Blanco Alcaine (Intel) Wednesday 4th March, 2020 Chair of Network Architectures and Services Department of Informatics Technical University of Munich

slide-2
SLIDE 2

Outline

  • Introduction to Time Sensitive Networking
  • Thesis Objectives
  • Design and Approach
  • Implementation
  • Network Calculus Model for WCD Analysis in TSN Networks
  • TSN emulation environment based on Mininet
  • GCL Synthesis Algorithms
  • TSN Testbed Setup
  • Evaluations
  • Conclusion and Future Works
  • A. Mildner — TSN

2

slide-3
SLIDE 3

Introduction to Time Sensitive Networking

Time Sensitive Networking (TSN)

  • A set of IEEE standards and additions to the IEEE 802.3 Ethernet standard, for providing

deterministic communication on common Ethernet technology (Layer 2)

  • IEEE Time Sensitive Networking Task Group, former Audio Video Bridging (AVB) Task

Group

  • High interest on TSN in various use-cases:
  • Industrial Automation (Industry 4.0)
  • Automotive
  • Aerospache/Avionic
  • Autonomous Driving
  • Convergence of Information Technology (IT) and Operational Technology (OT) possible

using TSN

  • The Standarization process is still ongoing, some important standards are not yet finished

and published

  • A. Mildner — TSN

3

slide-4
SLIDE 4

Introduction to Time Sensitive Networking

IEEE TSN Standards

Figure 1: TSN components [2]

  • A. Mildner — TSN

4

slide-5
SLIDE 5

Introduction to Time Sensitive Networking

Example TSN Network

ES1 ES2 ES3 ES4 SW1 SW2 Flow 1 Flow 2

GCL GCL GCL GCL GCL GCL

CUC CNC Figure 2: IEEE 802.1Qbv enabled TSN Network with a centralized Configuration Approach.

Legend:

  • CUC - Centralized User Configuration
  • CNC - Centralized Network Configuration
  • ESx - End Station
  • SWx - Switch
  • GCL - Gate Control List
  • A. Mildner — TSN

5

slide-6
SLIDE 6

Introduction to Time Sensitive Networking

IEEE 802.1Qbv - Time Aware Scheduler

Q7 Q6 Q5 Q0 Gate 7 Gate 6 Gate 5 Gate 0

Strict Prirority Scheduling Credit Based Shaper

Transmission Selection Algorithm Transmission Selection Algorithm

GCL T1 : 1000 0000 T2 : 0111 1111 TGCL Transmission Selection Switching Fabric egress ingress Figure 3: Time-based Scheduler (GCL: Gate Control List, TGCL : Cycle Time of the schedule)

  • A. Mildner — TSN

6

slide-7
SLIDE 7

Introduction to Time Sensitive Networking

IEEE 802.1Qbv - Time Aware Scheduler

Q7 Q6 Q5 Q0 Gate 7 Gate 6 Gate 5 Gate 0

Strict Prirority Scheduling Credit Based Shaper

Transmission Selection Algorithm Transmission Selection Algorithm

GCL T1 : 1000 0000 T2 : 0111 1111 TGCL Transmission Selection Switching Fabric egress ingress Figure 4: Time-based Scheduler (GCL: Gate Control List, TGCL : Cycle Time of the schedule)

  • A. Mildner — TSN

7

slide-8
SLIDE 8

Thesis Objectives

Problem Statement

  • The TSN set of standards is still lacking a proper way for dynamically (re-)configuration of

TSN networks

  • Currently TSN networks need to be statically configured with prior knowledge about the

network and the particular flows in it

  • The scheduling problem introduced by IEEE 802.1Qbv (Time Aware Scheduling) is con-

sidered as non-trival (NP-Complete) with respect to the network size

  • But: There have been recent research efforts, in order to tackle the dynamic schedule

synthesis Main goal of this Thesis: Provide and evaluate an apporach towards dynamic configuration and schedule synthesis for IEEE 802.1Qbv enabled TSN networks.

  • A. Mildner — TSN

8

slide-9
SLIDE 9

Thesis Objectives

  • Objective 1:

Provide an automated framework for worst-case delay analysis on a flow basis in IEEE 802.1Qbv scheduled TSN networks and evaluate and validate the implementation.

  • Objective 2:

Provide an automated framework for online GCL synthesis for IEEE 802.1Qbv based TSN networks and evaluate and validate the implementation.

  • Objective 3:

Provide suitable evaluation environments for automated measurements and validation for IEEE 802.1Qbv based TSN networks.

  • Objective 4:

Perform proper Analysis on the implemented methods for online schedule synthesis and dynamic configuration of IEEE 802.1Qbv enabled TSN networks.

  • A. Mildner — TSN

9

slide-10
SLIDE 10

Design and Approach

Dynamic Configuration System model design

Network Topology Stream Input Schedule Generator WCD Analysis Create Configuration Deploy Configuration User

Figure 5: Proposed system model design for Dynamic Configuration of IEEE 802.1Qbv enabled TSN networks.

  • Inputs: Network Topology, Stream Definitions
  • Output: Deployable Qbv configuration for the TSN network
  • The User can change the stream inputs i.e. add or remove streams or alter stream para-

meters

  • The generated GCLs should be validated using a NC Model for WCD Analysis for verifying,

that the end-to-end latency requirements are met

  • The generated configurations should be in the correct format for deployment in a TSN

network using the Deploy Configuration module

  • A. Mildner — TSN

10

slide-11
SLIDE 11

Implementation

Network Calculus Model for WCD Analysis in TSN Networks

  • L. Zaho et. al proposed in [5] a Network Calculus model for determining the Worst Case

Delay of flows in time-based scheduled TSN networks

  • Conducts a hop by hop WCD analysis, using leaky bucket arrival (α(t)) and TDMA service

curves (β(t))

  • Works on statically configured time-based scheduled TSN networks (GCLs given) and

accounts for impacts of higher and lower priority traffic

  • Inputs → network topology, GCLs, flow information, Flow of Interest (FOI)
  • Output → End-to-End WCD of the FOI
  • Implemented parameter calculation for the resulting service curves for each port on the

FOIs path in Python

  • For reproducable and reiable WCD analysis, we have used the DiscoDNC1 Framework

1 https://disco.cs.uni-kl.de/index.php/projects/disco-dnc

  • A. Mildner — TSN

11

slide-12
SLIDE 12

Implementation

Network Calculus Model for WCD Analysis in TSN Networks

Figure 6: Example resulting guaranteed service curve for a particular flow provided by a Qbv enabled Switch. [5]

  • A. Mildner — TSN

12

slide-13
SLIDE 13

Implementation

Building a TSN emulation environment based on Mininet

  • Nice to Have: A flexible Network Emulator that can do time-based scheduling according to

the standard

  • Idea: Use recently added TAPRIO2 kernel net-scheduler module to add time-based sched-

uling capabilities to mininet3

  • TAPRIO can be configured like any other Queuing Dicipline (qdisc) using the iproute2 tool

tc

  • Problem:
  • mininet uses veth (virtual ethernet) interfaces
  • veth interfaces implement no Transmit (TX) Queues
  • TAPRIO requires TX Queues to work
  • Result: Problem still persists and could not be solved during this thesis

2 http://man7.org/linux/man-pages/man8/tc-taprio.8.html 3 http://mininet.org/

  • A. Mildner — TSN

13

slide-14
SLIDE 14

Implementation

GCL Synthesis Algorithm

  • We implemented one example of a proposed GCL Synthesis Algorithm
  • Model is based on Array Theory Encoding (TA) for Satisfiability Modulo Theory (SMT) as

proposed in [4]

  • Implemented in Python, using the z3 SMT/OMT solver
  • Inputs → network topology, flow information, additional network information
  • Output → For each node in the network: array of open times (φ), array of close

times (τ) and array of frame-to-window assignment (κ)

  • Hide complexity of the underlying model behind a easy to use Class (GCLATSolver()) with

a sophisticated interface

  • Included a transformation function, in order to use the generated output with the real Hard-

ware on the TSN Testbed

  • A. Mildner — TSN

14

slide-15
SLIDE 15

Implementation

TSN Testbed Implementations

  • Idea: Conduct end-to-end latency measurements of a dynamically configurable time-based

scheduled TSN Network on real TSN capable Hardware

  • Setup a TSN Testbed consiting of 2 Qbv capable Switches and four to six TSN capable

End Stations at the Intel Office

  • We have implemented an example real-time application (talker and listener), which utilizes

the Intel i210 launchtime feature and Hardware TX/RX Timestamping

  • Implemented sophisticated configuration scripts:
  • End Stations: VLAN, TAPRIO and ETF configuration (remotely via SSH)
  • Switches: Qbv configuration using netopeer2 (NETCONF/RESTCONF Protocol)
  • A. Mildner — TSN

15

slide-16
SLIDE 16

Evaluation

NC model evaluation

  • For very simple scenarios the NC model seems to provide resonable results
  • We conducted a WCD Analysis on some of the simple example cases presented in the

paper [5]

  • One result was e.g. 1204.6µs (our implementation) versus 1287.8µs (paper)
  • Unfortunately, due to the lack of time we could not further investigate the cause of the

difference

  • A. Mildner — TSN

16

slide-17
SLIDE 17

Evaluation

Runtime Evaluation of the GCL synthesis Algorithm

#Streams 5 10 20 30 40 50 # O p e n W i n d

  • w

s 2 4 8 16 time [s] 50 100 150 200 250 300 350

15 Endstations | 5 Switches | Periods=['20ms', '10ms'] | 4 TT-Queues

100 200 300

Figure 7: Runtime results of the GCL synthesis Algorithm using Array Theory Encoding.

  • A. Mildner — TSN

17

slide-18
SLIDE 18

Evaluation

Runtime Evaluation of the GCL synthesis Algorithm

Figure 8: Scalability Analysis of the GCL synthesis Algorithm in a strict periodic system.

  • A. Mildner — TSN

18

slide-19
SLIDE 19

Evaluation

GCL Verification

20 40 60 80 100 TGCL [ s] ('ES1', 'SW1') ('SW1', 'ES3')

GCLs in the Network Figure 9: Generated GCLs in a simple TSN network setup with 2 ES and 1 SW.

  • A. Mildner — TSN

19

slide-20
SLIDE 20

Evaluation

TSN Testbed evaluation

launch txts [name] 10.3 10.4 10.5 10.6 10.7 10.8 10.9 11.0 E2E Latency in [ s]

E2E Latencies for 50000 samples.

Figure 10: ETF evaluation for an feasibility analysis of the generated Schedules.

  • Open window of 10µs, 400Byte packet size, one packet each 10ms
  • A. Mildner — TSN

20

slide-21
SLIDE 21

Conclusion and Future Works

Conclusion

  • We designed a system for dynamic configuration of IEEE 802.1Qbv enabled TSN networks

and implemented most of the proposed modules

  • The NC model for WCD Analysis on TSN networks showed resonable results for very

simple scenarios

  • The implemented GCL synthesis Algorithm showed good results for small networks, but

seems to be not suitable for larger networks

  • The latency measurements on the TSN testbed verified the feasibility of the generated

schedules But: There are still many tasks and future works left open

  • A. Mildner — TSN

21

slide-22
SLIDE 22

Conclusion and Future Works

Open Future Works

  • Enable TAPRIO on mininet (veth interfaces)
  • Refine the NC Model or implement another suggested NC model for Time Aware Sched-

uled netwowrks as e.g. presented in [3]

  • Implement and evaluate additional GCL synthesis Algorithms, e.g. as presented in [1] or

simpler heuristic apporaches

  • Complete the proposed system design, by putting together the already existing modules
  • Conduct measurements of the total re-configuration time of a dynamic configurable TSN

network

  • Implement a TSN Demonstrator, which utilizes and shows the dynamic configuration mech-

anisms

  • and more...

Thanks for your Attention !

  • A. Mildner — TSN

22

slide-23
SLIDE 23

Backup

TAPRIO configuration example tc qdisc add dev IFACE parent root handle 100 taprio num_tc 3 # Number of traffic Classes map 2 2 1 0 2 2 2 2 2 2 2 2 2 2 2 2 # Map Traffic Class -> SKB Priority queues 1@0 1@1 2@2 # Map Traffic Class -> HW Queue base-time 10000000 # Start Time sched-entry S 03 300000 # 1st Schedule Entry sched-entry S 02 300000 # 2nd Schedule Entry sched-entry S 06 400000 # 3rd Schedule Entry clockid CLOCK_TAI # Clock Source to use (Reference Clock)

  • A. Mildner — TSN

23

slide-24
SLIDE 24

Backup

Overlapping Scenarios in GCLs

  • A. Mildner — TSN

24

slide-25
SLIDE 25

Bibliography

[1]

  • F. Dürr and N. G. Nayak.

No-wait Packet Scheduling for IEEE Time-sensitive Networks (TSN). In Proceedings of the 24th International Conference on Real-Time Networks and Systems, RTNS ’16, pages 203–212, New York, NY, USA, 2016. ACM. [2]

  • J. Farkas.

Introduction to IEEE 802.1 - Focus on the Time-Sensitive Networking Task Group, 2017. http://www.ieee802.org/1/files/public/docs2017/tsn-farkas-intro-0517-v01.pdf. [3]

  • N. G. Nayak, F. Dürr, and K. Rothermel.

Routing algorithms for ieee802.1qbv networks. SIGBED Rev., 15(3):13–18, Aug. 2018. [4]

  • R. Serna Oliver, S. S. Craciunas, and W. Steiner.

IEEE 802.1Qbv Gate Control List Synthesis Using Array Theory Encoding. In 2018 IEEE Real-Time and Embedded Technology and Applications Symposium (RTAS), pages 13–24, April 2018. [5]

  • L. Zhao, P

. Pop, and S. S. Craciunas. Worst-Case Latency Analysis for IEEE 802.1Qbv Time Sensitive Networks Using Network Calculus. IEEE Access, 6:41803–41815, 2018.

  • A. Mildner — TSN

25