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

杨利霞

RouteNet: Leveraging Graph Neural Networks for

/ d% E0 l* M3 c# F2 Z6 `* q' e; C

Network Modeling and Optimization in SDN

7 f% p1 E! C# D, V9 q0 U5 {
) P: K3 v3 f! Y! W# A
' i2 X% T9 Q3 R, F4 ]1 rNetwork modeling is a key enabler to achieve
effificient network operation in future self-driving Software
1 x3 w& x! F2 K# fDefifined Networks. However, we still lack functional network
: A7 _' u- ?$ U6 W; Bmodels able to produce accurate predictions of Key Performance
5 V% s4 p# Y* m0 B$ v, CIndicators (KPI) such as delay, jitter or loss at limited cost.
) ~0 ~" P6 {  ~9 f6 h- A9 o6 lIn this paper we propose RouteNet, a novel network model based
on Graph Neural Network (GNN) that is able to understand
7 b: B' D3 s! L9 t& g8 ~+ L8 E3 ~the complex relationship between topology, routing, and input
traffific to produce accurate estimates of the per-source/destination
per-packet delay distribution and loss. RouteNet leverages the
" B! b: @, E4 V+ e$ g) ^: j4 uability of GNNs to learn and model graph-structured information
: Y" P3 v# s2 @0 z$ J6 ^9 L7 r  Iand as a result, our model is able to generalize over arbitrary
topologies, routing schemes and traffific intensity. In our eval
1 T: z3 z* h% y. y1 I: u' Z: uuation, we show that RouteNet is able to predict accurately
the delay distribution (mean delay and jitter) and loss even in
# ?! o! b/ c# o( ktopologies, routing and traffific unseen in the training (worst case
: Q# S& |/ A* ?7 zMRE = 15.4%). Also, we present several use cases where we
leverage the KPI predictions of our GNN model to achieve
0 u4 S* s9 Y/ k# D! f4 e- Meffificient routing optimization and network planning.


: h1 Q/ e/ ~% V