Gain More for Less: The Surprising Benefits of QoS Management in - - PowerPoint PPT Presentation

Gain More for Less: The Surprising Benefits of QoS Management in Constrained NDN Networks

ACM ICN 2019, Macau Cenk Gündoğan1 Jakob Pfender2 Michael Frey3 Thomas C. Schmidt1 Felix Shzu-Juraschek3 Matthias Wählisch4

1HAW Hamburg 2Victoria University of Wellington 3Safety IO 4Freie Universität Berlin

Common IoT Deployments

◮ Always connected, low-cost IoT devices

◮ Resource-constrained: MHz CPU, kB RAM/ROM

2 / 39

Common IoT Deployments

◮ Always connected, low-cost IoT devices

◮ Resource-constrained: MHz CPU, kB RAM/ROM

◮ Saturated resources impact network performance

◮ Local bottlenecks leave the network partially underutilized

3 / 39

Common IoT Deployments

◮ Always connected, low-cost IoT devices

◮ Resource-constrained: MHz CPU, kB RAM/ROM

◮ Saturated resources impact network performance

◮ Local bottlenecks leave the network partially underutilized

◮ Overprovisioning of resources to meet requirements ...

4 / 39

Common IoT Deployments

◮ Always connected, low-cost IoT devices

◮ Resource-constrained: MHz CPU, kB RAM/ROM

◮ Saturated resources impact network performance

◮ Local bottlenecks leave the network partially underutilized

◮ Overprovisioning of resources to meet requirements ... is infeasible

◮ Device complexity, unit price, and energy consumption increases

5 / 39

Common IoT Deployments

◮ Always connected, low-cost IoT devices

◮ Resource-constrained: MHz CPU, kB RAM/ROM

◮ Saturated resources impact network performance

◮ Local bottlenecks leave the network partially underutilized

◮ Overprovisioning of resources to meet requirements ... is infeasible

◮ Device complexity, unit price, and energy consumption increases

Quality of Service (QoS) improves resource utilization

6 / 39

Outline

Resources in IP vs. NDN Distributed QoS Management Experimental Evaluation Conclusion & Outlook

7 / 39

Resources in IP vs. NDN

◮ Typical IP world resources: link capacities & buffer spaces Forwarding Queues IP Resources

8 / 39

Resources in IP vs. NDN

◮ Typical IP world resources: link capacities & buffer spaces ◮ CCNx / NDN provides additional resources: Pending Interest Table (PIT), Content Store (CS) Forwarding Queues IP Resources PIT CS NDN Resources

9 / 39

Distributed QoS Management

QoS Building Blocks

• 1. Traffic classification
• 2. QoS treatments

11 / 39

QoS Building Blocks

• 1. Traffic classification

◮ Longest prefix match (LPM) with pre-defined name↔priority table ◮ Alternatively: draf-moiseenko-icnrg-flowclass, I. Moiseenko and D. Oran

• 2. QoS treatments

12 / 39

QoS Building Blocks

• 1. Traffic classification

◮ Longest prefix match (LPM) with pre-defined name↔priority table ◮ Alternatively: draf-moiseenko-icnrg-flowclass, I. Moiseenko and D. Oran

• 2. QoS treatments ⇐ focus of this talk

◮ Define quality dimensions ◮ Specify resource management rules

13 / 39

Quality Dimensions Latency Reliability Regular, Prompt Reliable, Prompt Regular, Regular Reliable, Regular

Toxic gas alerts in underground mines Temperature readings in a class room

14 / 39

Resource Management Rules

• 1. Isolated Decisions
• 2. Resource Correlations
• 3. Distributed Coordination

Forwarding Queue Delay regular traffic Pending Interest Table Evict regular for prompt Content Store Evict regular for reliable CS—PIT Correlation Prompt Data meets no PI ⇒ cached with priority CS—Forward. Correlation Prompt Data dropped ⇒ cached with priority PIT Coherence Same config. at all nodes

⇒ Regular < Reliable < Prompt

CS Efficiency Same config. at all nodes

⇒ Regular < Prompt < Reliable

15 / 39

Resource Management Rules

• 1. Isolated Decisions
• 2. Resource Correlations
• 3. Distributed Coordination

Forwarding Queue Delay regular traffic Pending Interest Table Evict regular for prompt Content Store Evict regular for reliable CS—PIT Correlation Prompt Data meets no PI ⇒ cached with priority CS—Forward. Correlation Prompt Data dropped ⇒ cached with priority PIT Coherence Same config. at all nodes

⇒ Regular < Reliable < Prompt

CS Efficiency Same config. at all nodes

⇒ Regular < Prompt < Reliable

16 / 39

Resource Management Rules

• 1. Isolated Decisions
• 2. Resource Correlations
• 3. Distributed Coordination

Forwarding Queue Delay regular traffic Pending Interest Table Evict regular for prompt Content Store Evict regular for reliable CS—PIT Correlation Prompt Data meets no PI ⇒ cached with priority CS—Forward. Correlation Prompt Data dropped ⇒ cached with priority PIT Coherence Same config. at all nodes

⇒ Regular < Reliable < Prompt

CS Efficiency Same config. at all nodes

⇒ Regular < Prompt < Reliable

17 / 39

Resource Management Rules

• 1. Isolated Decisions
• 2. Resource Correlations
• 3. Distributed Coordination

Forwarding Queue Delay regular traffic Pending Interest Table Evict regular for prompt Content Store Evict regular for reliable CS—PIT Correlation Prompt Data meets no PI ⇒ cached with priority CS—Forward. Correlation Prompt Data dropped ⇒ cached with priority PIT Coherence Same config. at all nodes

⇒ Regular < Reliable < Prompt

CS Efficiency Same config. at all nodes

⇒ Regular < Prompt < Reliable

18 / 39

Experimental Evaluation

Experimental Evaluation Setup

Hardware: M3 Node in IoT Lab testbed Sofware: RIOT with CCN-lite Network: Multi-hop topology with 31 nodes

Gateway

M3 Node (ARM Cortex-M3) 64 kB RAM / 512 kB ROM 802.15.4 radio transceiver

20 / 39

Scenario Descriptions

Mixed Sensors and Actuators Sensing and Lighting Control

21 / 39

Scenario Descriptions

Mixed Sensors and Actuators

◮ Gateway requests device-specific temperature readings every 10 s ± 2 s

Sensing and Lighting Control

Interest

Gateway Traffic

22 / 39

Scenario Descriptions

Mixed Sensors and Actuators

◮ Gateway requests device-specific temperature readings every 10 s ± 2 s ◮ Actuators request device-specific state from gateway every 5 s ± 1 s

Sensing and Lighting Control

Interest

Gateway Traffic

Interest

Actuators Traffic

23 / 39

Scenario Descriptions

Mixed Sensors and Actuators

◮ Gateway requests device-specific temperature readings every 10 s ± 2 s ◮ Actuators request device-specific state from gateway every 5 s ± 1 s

Sensing and Lighting Control

◮ Actuators request group-specific instructions from gateway every 5 s ± 1 s

Interest

Gateway Traffic

Interest

Actuators Traffic

cache hit cache hit 24 / 39

Evaluation Metrics

Success Throughput Latency

25 / 39

Evaluation Metrics: Success Rates

Success

Throughput Latency

26 / 39

Nodal Success Rates for Actuators Traffic

20 40 60 80 100

Success Rate [%] Regular

27 / 39

Nodal Success Rates for Actuators Traffic

20 40 60 80 100

Success Rate [%] Regular QoS Coordinated

28 / 39

Overall Success Rates

2 4 6 8 10 12 Regular 20 40 60 80 100 Success Rate [%] QoS Coordinated Regular QoS Coordinated 5 10 15 20 25 30 2 4 6 8 10 12 5 10 15 20 25 30 5 10 15 20 25 30 5 10 15 20 25 30 PIT Size [Maximum # of Entries] Mixed Sensors & Actuators CS Size [Maximum # of Entries] Sensing & Lighting Control Rank [Hops]

Actuators Gateway

29 / 39

Overall Success Rates

2 4 6 8 10 12 Regular 20 40 60 80 100 Success Rate [%] QoS Coordinated Regular QoS Coordinated 5 10 15 20 25 30 2 4 6 8 10 12 5 10 15 20 25 30 5 10 15 20 25 30 5 10 15 20 25 30 PIT Size [Maximum # of Entries] Mixed Sensors & Actuators CS Size [Maximum # of Entries] Sensing & Lighting Control Rank [Hops]

Actuators Gateway

30 / 39

Overall Success Rates

2 4 6 8 10 12 Regular 20 40 60 80 100 Success Rate [%] QoS Coordinated Regular QoS Coordinated 5 10 15 20 25 30 2 4 6 8 10 12 5 10 15 20 25 30 5 10 15 20 25 30 5 10 15 20 25 30 PIT Size [Maximum # of Entries] Mixed Sensors & Actuators CS Size [Maximum # of Entries] Sensing & Lighting Control Rank [Hops]

Actuators Gateway

CS5 CS15 CS30 20 40 CS Size [Maximum # of Entries] Cache Hit [%] Regular QoS Coordinated

31 / 39

Evaluation Metrics: Throughput Evolution

Success

Throughput

Latency

32 / 39

Throughput Evolution for Unprioritized Traffic

w/o Actuator Traffic w/ Actuator Traffic

200 400 600 Regular

w/o Actuator Traffic w/ Actuator Traffic

QoS Coordinated

w/o Actuator Traffic w/ Actuator Traffic

5 10 15 20 200 400 600

w/o Actuator Traffic w/ Actuator Traffic

5 10 15 20 Duration [min] Packets [ #

min]

Outgoing Interests Incoming Data

33 / 39

Throughput Evolution for Unprioritized Traffic

w/o Actuator Traffic w/ Actuator Traffic

200 400 600 Regular

w/o Actuator Traffic w/ Actuator Traffic

QoS Coordinated

w/o Actuator Traffic w/ Actuator Traffic

5 10 15 20 200 400 600

w/o Actuator Traffic w/ Actuator Traffic

5 10 15 20 Duration [min] Packets [ #

min]

Outgoing Interests Incoming Data

34 / 39

Goodput Evolution

5 10 15 2 4 6

Regular

PIT5 PIT30

5 10 15

QoS Coordinated

PIT5 PIT30

2 4 6 8 10 12 2 4 6 8 10 12 0.2 0.4 0.6

PIT5 PIT30

2 4 6 8 10 12 2 4 6 8 10 12

Goodput [ KiB

min]

Rank [Hops] Duration [min]

Gateway Actuators

35 / 39

Goodput Evolution

5 10 15 2 4 6

Regular

PIT5 PIT30

5 10 15

QoS Coordinated

PIT5 PIT30

2 4 6 8 10 12 2 4 6 8 10 12 0.2 0.4 0.6

PIT5 PIT30

2 4 6 8 10 12 2 4 6 8 10 12

Goodput [ KiB

min]

Rank [Hops] Duration [min]

Gateway Actuators

36 / 39

Evaluation Metrics: Completion Time

Success Throughput

Latency

37 / 39

Nodal Completion Time for Actuators Traffic

2 4 6 8 1012 14 1618202224262830

200 400 Time to Completion [ms] Regular

2 4 6 8 1012 14 1618202224262830

Reliable

2 4 6 8 1012 14 1618202224262830

Prompt Nodes sorted by hop distance to gateway [NodeID] ← closer farther →

38 / 39

Conclusion & Outlook

Takeaways

◮ PIT and cache space have prevailing effects on overall network performance ◮ QoS in NDN is not confined to simple resource trading ◮ Treating Interest as well as Data messages allows for resource correlations ◮ Unprioritized traffic benefits from resource coordination

Next Steps

◮ Investigate further correlations between PIT, CS, and buffer spaces ◮ Elaborate on the choice of quality dimensions and service levels

39 / 39