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

杨利霞

RouteNet: Leveraging Graph Neural Networks for

( v+ f( x$ J; H. C( k

Network Modeling and Optimization in SDN

Network modeling is a key enabler to achieve
. p0 Z! f+ i3 q" F. S' K* \effificient network operation in future self-driving Software
Defifined Networks. However, we still lack functional network
5 a( y# z8 z( x4 o( K: j' ~0 @1 Hmodels able to produce accurate predictions of Key Performance
6 t* S% y/ O& [* ^+ B2 q4 OIndicators (KPI) such as delay, jitter or loss at limited cost.
In this paper we propose RouteNet, a novel network model based
7 i% J% y. s2 [9 }  ]" M: ^on Graph Neural Network (GNN) that is able to understand
* q" t9 g( D/ B' R" B; Cthe complex relationship between topology, routing, and input
# V3 t" G/ z0 Y9 v5 B0 Ttraffific to produce accurate estimates of the per-source/destination
per-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
topologies, routing schemes and traffific intensity. In our eval
# ?% G4 |6 {4 Q- u; f% Uuation, we show that RouteNet is able to predict accurately
, j6 w$ Y+ b( g6 Pthe delay distribution (mean delay and jitter) and loss even in
topologies, routing and traffific unseen in the training (worst case
MRE = 15.4%). Also, we present several use cases where we
9 U* b* f5 s+ \/ j, Bleverage the KPI predictions of our GNN model to achieve
effificient routing optimization and network planning.