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

杨利霞

RouteNet: Leveraging Graph Neural Networks for

3 S% I! u# I" _

Network Modeling and Optimization in SDN

# }0 O4 {' Y. O
% `$ ^9 p. d; V' s" F* ^+ I" BNetwork modeling is a key enabler to achieve
effificient network operation in future self-driving Software
" H" x( G( U$ W: C; DDefifined Networks. However, we still lack functional network
5 t  J8 ^2 c! E" mmodels able to produce accurate predictions of Key Performance
/ j7 R' r7 E3 D4 f6 _3 mIndicators (KPI) such as delay, jitter or loss at limited cost.
In this paper we propose RouteNet, a novel network model based
, a. h/ i% k- Z  E0 E& hon Graph Neural Network (GNN) that is able to understand
the complex relationship between topology, routing, and input
traffific to produce accurate estimates of the per-source/destination
' _9 r' Z& K6 Q' vper-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
2 F9 q0 C5 m1 f2 v/ D" X1 k( otopologies, routing schemes and traffific intensity. In our eval
uation, we show that RouteNet is able to predict accurately
the delay distribution (mean delay and jitter) and loss even in
1 [' Y! O7 C" n9 O6 H8 A/ Ltopologies, routing and traffific unseen in the training (worst case
MRE = 15.4%). Also, we present several use cases where we
: s# a( J7 z* F$ [) p5 @9 Lleverage the KPI predictions of our GNN model to achieve
4 W. G# @/ H5 z  G' }effificient routing optimization and network planning.
% e  q# `/ Y) n1 x# e0 J7 z. J  i, i

+ r" M% u5 b) S1 t. {