11
Traffic Forecasting Medium Access TRANSFORMA Vladislav Petkov Katia Obraczka 1

Traffic Forecasting Medium Access TRANSFORMA Vladislav Petkov Katia Obraczka 1

  • View
    215

  • Download
    1

Embed Size (px)

Citation preview

Page 1: Traffic Forecasting Medium Access TRANSFORMA Vladislav Petkov Katia Obraczka 1

Traffic Forecasting Medium Access

TRANSFORMA

Vladislav PetkovKatia Obraczka

1

Page 2: Traffic Forecasting Medium Access TRANSFORMA Vladislav Petkov Katia Obraczka 1

Motivation• Goals of a good wireless MAC protocol– High channel utilization– Predictable performance under heavy load– Energy efficiency– Low latency

• Schedule based MACs provide first three at expense of the fourth

• We want to address that problem• Using per-flow traffic forecasting:– Can determine rate of service flow needs– Allocate just right amount of resources to each flow

2

Page 3: Traffic Forecasting Medium Access TRANSFORMA Vladislav Petkov Katia Obraczka 1

Related work

• Other schedule based approaches:– Traffic Adaptive Medium Access (TRAMA)– Flow Aware Medium Access (FLAMA)– Dynamic Multi-Channel Medium Access

(DYNAMMA)

3

Page 4: Traffic Forecasting Medium Access TRANSFORMA Vladislav Petkov Katia Obraczka 1

TRANSFORMA design

• Time slotted channel

• 2-hop neighborhood information propagated

• Distributed medium scheduling

4

Page 5: Traffic Forecasting Medium Access TRANSFORMA Vladislav Petkov Katia Obraczka 1

Control Plane

5

Page 6: Traffic Forecasting Medium Access TRANSFORMA Vladislav Petkov Katia Obraczka 1

Traffic forecasterFor every packet arrival traffic forecaster runs following algorithm:1.Computes the latest packet interval, τ2.Calculates loss of each expert, xi

3.Reduce weights of bad experts4.Share some of remaining weights5.Calculate new forecaster output

6

Page 7: Traffic Forecasting Medium Access TRANSFORMA Vladislav Petkov Katia Obraczka 1

Flow selection algorithm• Rate Monotonic Algorithm

approximated to select flow for each slot– Flow fi, with interval ti,

pseudorandomly chooses one slot in each ti interval

– Ties are broken by flow and node ID hash

– Flows persist (try for every slot) if they fail to win their chosen slot

• Flows with interval below τmin are considered best-effort

• We empirically determine best τmin value (currently 10ms)

• TRANSFORMA guarantees collision free medium access under all conditions

7

Page 8: Traffic Forecasting Medium Access TRANSFORMA Vladislav Petkov Katia Obraczka 1

Performance evaluation

• We use Qualnet Network Simulator

• PHY is 802.11a at 6.0Mbps• Radio range is ≈400m• 2 experiments:

– Heterogeneous flows• Variable number of CBR

flows of different rates

– Real-time vs. best-effort• 3 real-time flows with

increasing amount of background traffic

Hotspot layout

8

Page 9: Traffic Forecasting Medium Access TRANSFORMA Vladislav Petkov Katia Obraczka 1

Expt 1 results: Heterogeneous Flows

TRANSFORMADYNAMMA (a schedule based protocol)

9

Page 10: Traffic Forecasting Medium Access TRANSFORMA Vladislav Petkov Katia Obraczka 1

Expt 2 results: Real-time vs best-effort

Foreground application delayForeground & background goodput

10

Page 11: Traffic Forecasting Medium Access TRANSFORMA Vladislav Petkov Katia Obraczka 1

Conclusion and future work

• TRANSFORMA has predictable performance, even at high load

• TRANSFORMA delivers lower delays to delay sensitive applications than DYNAMMA and even 802.11 under high load

• Future work:– Implementation based experimentation with

more applications– Ways of adding routing awareness in scheduling

11