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

杨利霞

RouteNet: Leveraging Graph Neural Networks for

Network Modeling and Optimization in SDN

8 ^  {. c5 w& t3 a* Q

Network modeling is a key enabler to achieve
effificient network operation in future self-driving Software
" i2 A: P; ^* q& _, w% cDefifined Networks. However, we still lack functional network
5 I& q% U6 |+ f5 M( X5 i: nmodels able to produce accurate predictions of Key Performance
0 V% l6 s; o; ^; QIndicators (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
% ?* k1 V( T" i' {the complex relationship between topology, routing, and input
# Q* b) o2 m/ G8 T1 x# B, ztraffific to produce accurate estimates of the per-source/destination
per-packet delay distribution and loss. RouteNet leverages the
( k' i8 s! L. p- m% e% u4 h/ }7 [ability of GNNs to learn and model graph-structured information
and as a result, our model is able to generalize over arbitrary
. B2 N2 b( J) Y6 Dtopologies, 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
topologies, routing and traffific unseen in the training (worst case
7 x" v$ p; P; j9 [1 b. N, GMRE = 15.4%). Also, we present several use cases where we
leverage the KPI predictions of our GNN model to achieve
3 K% H- L5 ]; G  Feffificient routing optimization and network planning.
* g4 {; i0 F* j4 M9 N
4 P% C, R; _$ x$ X1 `( o6 @2 y