DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks Fabien Geyer 1,2 and Steffen Bondorf 3 INFOCOM 2019 Wednesday 1 st May, 2019 1 Chair of Network Architectures and Services 3 Dept. of Information Security and Communication Technology 2 Airbus Central R&T Technical University of Munich (TUM) Munich Norwegian University of Science and Technology (NTNU)
Motivation Worst-Case End-to-End Performance Analysis Probability Deadline • Important for critical applications • Need formal proof on network delay Worst-case End-to-end network delay F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Worst-Case End-to-End Performance Analysis Computation effort Probability Deadline Analysis methods Worst-case Bound Measurements Tightness Simulation Ideal End-to-end network delay Tightness improvement • Trade-off between computational effort and tightness • This talk: network analysis method with good tightness and fast execution F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Network Calculus – Basics Basis: Cumulative arrivals and services [Cruz, 1991a] Data A D Time F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Network Calculus – Basics Basis: Cumulative arrivals and services [Cruz, 1991a] Data A D A Time F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Network Calculus – Basics Basis: Cumulative arrivals and services [Cruz, 1991a] Data A D A D Time F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Network Calculus – Basics Basis: Cumulative arrivals and services [Cruz, 1991a] Data A D A D Backlog Time F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Network Calculus – Basics Basis: Cumulative arrivals and services [Cruz, 1991a] Data A D A D Delay Time F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Network Calculus – Basics Basis: Cumulative arrivals and services [Cruz, 1991a] Data A D A Arrival curve α : A ( t ) − A ( t − s ) ≤ α ( s ), ∀ t ≤ s D α Time F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Network Calculus – Basics Basis: Cumulative arrivals and services [Cruz, 1991a] Data A D A Arrival curve α : A ( t ) − A ( t − s ) ≤ α ( s ), ∀ t ≤ s D α Time F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Network Calculus – Basics Basis: Cumulative arrivals and services [Cruz, 1991a] Data A D A Arrival curve α : A ( t ) − A ( t − s ) ≤ α ( s ), ∀ t ≤ s D α Time F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Network Calculus – Basics Basis: Cumulative arrivals and services [Cruz, 1991a] Data A D A Arrival curve α : A ( t ) − A ( t − s ) ≤ α ( s ), ∀ t ≤ s D Service curve β : a server is said to offer a strict service curve β if, during any backlogged period of duration u , the output of the system is at least equal to α β ( u ) β Time F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Network Calculus – Basics Basis: Cumulative arrivals and services [Cruz, 1991a] Data A D A Arrival curve α : A ( t ) − A ( t − s ) ≤ α ( s ), ∀ t ≤ s D Service curve β : a server is said to offer a strict service curve β if, during any backlogged period of duration u , the output of the system is at least equal to α β ( u ) β Time F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Network Calculus – Basics Basis: Cumulative arrivals and services [Cruz, 1991a] Data A D A Arrival curve α : A ( t ) − A ( t − s ) ≤ α ( s ), ∀ t ≤ s D Service curve β : a server is said to offer a strict service curve β if, during any backlogged period of duration u , the output of the system is at least equal to α β ( u ) β Time F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Network Calculus – Basics Basis: Cumulative arrivals and services [Cruz, 1991a] Data A D Arrival curve α : A ( t ) − A ( t − s ) ≤ α ( s ), ∀ t ≤ s Service curve β : a server is said to offer a strict service curve β if, during any backlogged period of duration u , the output of the system is at least equal to α β ( u ) ′ α α β β Time F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Network Calculus – Basics Basis: Cumulative arrivals and services [Cruz, 1991a] Data A D Arrival curve α : A ( t ) − A ( t − s ) ≤ α ( s ), ∀ t ≤ s Service curve β : a server is said to offer a strict Backlog Bound service curve β if, during any backlogged period of duration u , the output of the system is at least equal to α β ( u ) ′ α α β β Time F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Network Calculus – Basics Basis: Cumulative arrivals and services [Cruz, 1991a] Data A D Arrival curve α : A ( t ) − A ( t − s ) ≤ α ( s ), ∀ t ≤ s Service curve β : a server is said to offer a strict service curve β if, during any backlogged period of duration u , the output of the system is at least equal to α β ( u ) Delay Bound ′ α α β β Time F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Network Calculus – Basics Basis: Cumulative arrivals and services [Cruz, 1991a] Data A D Arrival curve α : A ( t ) − A ( t − s ) ≤ α ( s ), ∀ t ≤ s Service curve β : a server is said to offer a strict service curve β if, during any backlogged period of duration u , the output of the system is at least equal to α β ( u ) ′ α α β β Time F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Network Calculus – Basics Basis: Cumulative arrivals and services [Cruz, 1991a] Data A D Arrival curve α : A ( t ) − A ( t − s ) ≤ α ( s ), ∀ t ≤ s Service curve β : a server is said to offer a strict service curve β if, during any backlogged period of duration u , the output of the system is at least equal to α β ( u ) ′ α α β β Time F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
Motivation Network Calculus – Basics Basis: Cumulative arrivals and services [Cruz, 1991a] Data A D ′ α Arrival curve α : A ( t ) − A ( t − s ) ≤ α ( s ), ∀ t ≤ s Service curve β : a server is said to offer a strict service curve β if, during any backlogged period of duration u , the output of the system is at least equal to β ( u ) ′ α α β Time F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 2
MODELING AND ANALYSIS OF NETWORK INFRASTRUCTURE IN CYBER-PHYSICAL SYSTEMS LIANG CHENG (LEHIGH UNIVERSITY, BETHLEHEM, USA) STEFFEN BONDORF (NTNU TRONDHEIM, NORWAY) ACM SIGCOMM 2019 TUTORIALS 2019-08-23 BEIJING, CHINA
Motivation Network Calculus – Network Analysis How to compute end-to-end performance? f 4 f 1 s 1 s 2 s 3 s 4 f 2 f 3 F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 3
Motivation Network Calculus – Network Analysis How to compute end-to-end performance? f 4 f 1 s 1 s 2 s 3 s 4 f 2 f 3 TFA – Total Flow Analysis [Cruz, 1991b] Step 1: Compute delay at each server on the path f 4 f 1 s 1 s 2 s 3 s 4 f 2 f s 3 f s 4 3 3 Step 2: Sum delays F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 3
Motivation Network Calculus – Network Analysis How to compute end-to-end performance? f 4 Server concatenation [Le Boudec and Thiran, 2001] f 1 s 1 s 2 s 3 s 4 ′ α α f 2 f 3 β 1 β 2 β 3 TFA – Total Flow Analysis [Cruz, 1991b] (min, +) algebra gives us: Step 1: Compute delay at each server on the path f 4 ′ α α f 1 β 1 ⊗ β 2 ⊗ β 3 s 1 s 2 s 3 s 4 f 2 f s 3 f s 4 3 3 → Pay Bursts Only Once principle Step 2: Sum delays F. Geyer and S. Bondorf — DeepTMA: Predicting Effective Contention Models for Network Calculus using Graph Neural Networks 3
Recommend
More recommend