Agenda Motivation Multipath approaches Modeling Experiments and - - PDF document

Institute of Computer Science Chair of Communication Systems

• Prof. Dr.-Ing. P. Tran-Gia

Performance of Concurrent Multipath Transmissions

• Measurements and Model Validation

Thomas Zinner, Dominik Klein, Kurt Tutschku, Tanja Zseby, Phuoc Tran-Gia, Yuval Shavitt

SJDE - MultiNext

7th EURO-NF CONFERENCE ON NEXT GENERATION INTERNET NGI 2011

2

Thomas Zinner

Agenda

Motivation Multipath approaches Modeling Experiments and results Conclusion

3

Thomas Zinner

MOTIVATION

4

Thomas Zinner

Motivation

Concurrent Multipath Transmission Resource pooling ! Parallel usage of resources ! Forming a virtualized single "transport resource# Aiming for Improved resilience High capacity (e.g. for video streaming) New business models e.g. ! Combine multiple operators ! Dynamic resource selection

5

Thomas Zinner

Concurrent Multipath Transfer

Superset of available physical resources Routing topologies

• f provider I

Routing topology

• f provider II

Aim: Very high and reliable throughput between two end hosts pooled transport pipe

POP Logical topology Aim: Very high and reliable transmission between two end hosts Solution: Transport Virtualization: Combine multiple paths (even from different providers) Remark: Multihomed end user devices (e.g. Smartphones, Notebooks,$) in general also allow multipath transmissions

6

Thomas Zinner

Packets arrive out of order at the destination Re-sequencing necessary How can we calculate the ratio of reordered packets (and then do proper dimensioning)?

Performance Issues

• ut-of-order arrivals

different path delays re-sequencing ?

7

Thomas Zinner

MULTIPATH APPROACHES

8

Thomas Zinner

Multipath Approaches

Current multipath transmission approaches Transport virtualization (Tutschku et al.) Multipath-TCP (Trilogy/IETF) Concurrent multipath transfer using SCTP (Iyengar et al.) Load balancing in MPLS (X. Hesselbach et al.) pTCP (Hsieh et al.) Reordering in OBS networks (Gunreben et. al) $ Common problems: out-of-order arrivals Our solution: re-sequencing buffer Buffer dimensioning (metric for out-of-order packets) Influence on the E2E delay and throughput of pooled resources Selection strategies

9

Thomas Zinner

MODELING

10

Thomas Zinner

Withpaths

Transmission Model

p1,1

dst

Assumption:usek parallelpathsonm

• verlays

p1,n1 pm,1 pm,nm

src

k paths

• 1

Datastreamdividedatingressinto segmentswithk parts 1 k 2 k partshave arrived k partsaresendin parallelattimet k1 eachproviderwillofferasetni ofparallelpaths (i =1 m)

• 1

k

• verlay1
• verlaym

Buffer occupancy?

Reassembledatastreamfrom

• btainedparts

Resequencingbuffer

• fsizeL

Scheduling?

Datastream

11

Thomas Zinner

Performance Parameters

Input Number of paths Scheduling Output: Re-sequencing buffer occupancy distribution E2E delay distribution Path delay distributions Path capacity

12

Thomas Zinner

Each time unit, a packet is transmitted over each path Constraint: packets do not overtake each other on a single path, i.e. current delay previous delay ! interdeparture time But: packets overtake each other on other paths:

Assumptions

path 1 destination path 2 packets in re-sequencing buffer

2 3 1 5 4 6 2 2 4 4 2 1 4 3 6 5

$ $

time

13

Thomas Zinner

Methodology

Analytical approach Buffer occupancy probability can be calculated by combinatorial analysis (cf. Nebat 2002, ITC 22): Simulative approach (ITC-Specialist Seminar 20) Autoregressive packet delay computation Scheduling mechanisms / E2E delay Measurements within experimental facilities Validation of analytical and simulative approach Limitations of theoretical models

P (B k )

i 1 m

• P (B k ,sn i ,i x )

x 0

• 14

Thomas Zinner

EXPERIMENTS AND RESULTS

15

Thomas Zinner

Measurement Setup

ETOMIC boxes at end points (Pamplona, Elte) Clock synchronization via GPS DAG cards (timestamps with precision of nano seconds) Application layer overlay enables multipath transmission Different multipath transmission scenarios ( Brazil ! Ireland, Brazil ! Brazil, Canada ! China) 14 experiments for each scenario (100k packets via each path)

16

Thomas Zinner

Path Delays

One way delay measurements: Ireland - Brazil

17

Thomas Zinner

Validation – Re-sequencing Buffer

Re-sequencing buffer for scenario Brazil ! Ireland Best and worst match between theoretical models and measurements (MSE) Good match between theory and experiments

18

Thomas Zinner

Validation – Re-sequencing Buffer

Re-sequencing buffer for scenario Brazil ! Brazil Good match for low path delay correlation Analytical model overestimates the buffer in case of a high path delay correlation (cross correlation > 0.7) Limited model accuracy for correlated path delays

19

Thomas Zinner

Investigation – E2E Delay

E2E delay for the scenario Brazil ! Ireland E2E delay dominated by path with higher delay (waiting time in the buffer negligible) Close match between simulation and measurements

20

Thomas Zinner

CONCLUSION

21

Thomas Zinner

Conclusion

Introduction of concurrent multipath transmissions as a resource pooling concept that may improve capacity and reliability Re-sequencing buffer occupancy and E2E delay as performance metrics Validation of theoretical models with measurements Good match of theoretical models and measurements in general Overestimation of analytical model for positive delay correlations E2E delay dominated by the path with higher delay Future Work: deep investigation of the system with theoretical models Appropriate scheduling mechanisms Impact of delay correlation Trade-off between E2E delay and buffer occupancy

22

Thomas Zinner

Questions?

Thank you%