数学建模社区-数学中国

标题: RouteNet: Leveraging Graph Neural Networks for Network Modeling and Optimizat... [打印本页]
作者: 杨利霞    时间: 2020-11-16 15:20
标题: RouteNet: Leveraging Graph Neural Networks for Network Modeling and Optimizat...
RouteNet: Leveraging Graph Neural Networks for
) t! k5 `1 a( B
Network Modeling and Optimization in SDN
* ^- z1 [2 f! X

' w9 v! W7 i% x* h" r- J8 I
3 Q# h- P4 J% \1 H; w8 m# `Network modeling is a key enabler to achieve
8 g) K1 Z6 a/ neffificient network operation in future self-driving Software- {9 h* B# f7 o$ d
Defifined Networks. However, we still lack functional network/ [6 y  y% F9 |1 v# l
models able to produce accurate predictions of Key Performance
/ K, }9 x5 e( MIndicators (KPI) such as delay, jitter or loss at limited cost.
6 m- _  X- }$ i  N8 r7 jIn this paper we propose RouteNet, a novel network model based- A) e5 M3 ~4 `2 f6 c
on Graph Neural Network (GNN) that is able to understand
# s% n: ^3 S! L1 dthe complex relationship between topology, routing, and input3 \. g  D) L3 Z. \# r$ q
traffific to produce accurate estimates of the per-source/destination
- L, _4 \# [( S  a& W5 Sper-packet delay distribution and loss. RouteNet leverages the
. `# w2 R6 J& X- G+ }1 H  Y) |ability of GNNs to learn and model graph-structured information
7 _, ^8 @$ Q3 _and as a result, our model is able to generalize over arbitrary  [9 r- k8 f  `+ ^' {! E& `# b" J
topologies, routing schemes and traffific intensity. In our eval
. r. Z% @! n- S+ Q- Vuation, we show that RouteNet is able to predict accurately* V2 D$ }$ ^& b+ }. z3 P
the delay distribution (mean delay and jitter) and loss even in1 M8 {  N, s0 \: \. g
topologies, routing and traffific unseen in the training (worst case
7 Y8 i9 c2 C2 Q0 }MRE = 15.4%). Also, we present several use cases where we
0 h& M- Y( Y  H/ \( Z/ U: [leverage the KPI predictions of our GNN model to achieve
: |$ B. R* U0 f, peffificient routing optimization and network planning.
  g" H. M6 z( o# E5 l3 \* Z* Z- C2 q6 C" x
  b8 K5 J' n+ G* h6 P

08934670.pdf

999 KB, 下载次数: 0, 下载积分: 体力 -2 点



欢迎光临 数学建模社区-数学中国 (http://www.madio.net/) Powered by Discuz! X2.5