RouteNet: Leveraging Graph Neural Networks for Network Modeling and Optimizat...

杨利霞

RouteNet: Leveraging Graph Neural Networks for

/ v5 _9 e, d# i9 \

Network Modeling and Optimization in SDN

/ P- j/ W- S2 S9 B6 |1 ^  [

Network modeling is a key enabler to achieve
- v0 \. r% L  R0 [2 Geffificient network operation in future self-driving Software
3 b7 X+ [+ P, oDefifined Networks. However, we still lack functional network
models able to produce accurate predictions of Key Performance
; k0 E) Z  f+ E: B6 DIndicators (KPI) such as delay, jitter or loss at limited cost.
In this paper we propose RouteNet, a novel network model based
  B& R0 J! I+ }on Graph Neural Network (GNN) that is able to understand
the complex relationship between topology, routing, and input
) G! Q" `/ b$ f* xtraffific to produce accurate estimates of the per-source/destination
per-packet delay distribution and loss. RouteNet leverages the
: z5 D2 y: J; U1 D& K% Nability of GNNs to learn and model graph-structured information
and as a result, our model is able to generalize over arbitrary
6 H5 y5 ~2 e: u8 htopologies, routing schemes and traffific intensity. In our eval
( G4 o6 x* k% `, g: K: ?! j# p; ]4 ouation, we show that RouteNet is able to predict accurately
; x* f6 g8 m% z8 [( `5 Ythe delay distribution (mean delay and jitter) and loss even in
) o' N; {2 F5 W) `+ d/ b4 I3 ?topologies, routing and traffific unseen in the training (worst case
% i. Q: ^* \( F: z3 @MRE = 15.4%). Also, we present several use cases where we
leverage the KPI predictions of our GNN model to achieve
effificient routing optimization and network planning.