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

杨利霞

RouteNet: Leveraging Graph Neural Networks for

6 Y' k2 F7 k0 C8 R* r6 d" c

Network Modeling and Optimization in SDN

2 S! l1 {. W! G2 Q0 b
Network modeling is a key enabler to achieve
7 O+ R( s5 V, T$ {9 Heffificient network operation in future self-driving Software
7 F3 y, x, z' f3 y1 UDefifined Networks. However, we still lack functional network
" u: e7 H3 T" e) b; O- a: dmodels able to produce accurate predictions of Key Performance
% Z7 P: T; }% j0 h" S! v  KIndicators (KPI) such as delay, jitter or loss at limited cost.
In this paper we propose RouteNet, a novel network model based
on Graph Neural Network (GNN) that is able to understand
the complex relationship between topology, routing, and input
, ]+ o. C$ f  X) }) v! z/ btraffific to produce accurate estimates of the per-source/destination
9 L$ U5 Y* w" pper-packet delay distribution and loss. RouteNet leverages the
ability of GNNs to learn and model graph-structured information
and as a result, our model is able to generalize over arbitrary
1 M: z9 x& e  |6 G2 ~topologies, routing schemes and traffific intensity. In our eval
( Q5 D( b. n! v! f: q' Juation, we show that RouteNet is able to predict accurately
the delay distribution (mean delay and jitter) and loss even in
, J$ C1 B  p  `* J$ @$ btopologies, routing and traffific unseen in the training (worst case
6 x8 D" J2 k9 H" K8 gMRE = 15.4%). Also, we present several use cases where we
leverage the KPI predictions of our GNN model to achieve
% H6 U; y# o' S/ eeffificient routing optimization and network planning.
7 a, c( |/ E/ u% g8 O9 b
7 Z- U/ J7 f" q) c- n
9 K. G& O. e" f# U