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
' h& n% h/ r$ W; V. J+ Q4 S
Network Modeling and Optimization in SDN
$ e* i, Q5 n/ k t( E
. m! | @5 J" Q) _! g5 g, n2 O
$ w1 J9 f) }5 s1 yNetwork modeling is a key enabler to achieve
$ I$ [3 R0 M J* i* c( [* b7 U" xeffificient network operation in future self-driving Software) U. w/ I ? q/ b8 R* u# c( D( Q
Defifined Networks. However, we still lack functional network' i8 F% a9 x: J. k! C
models able to produce accurate predictions of Key Performance
+ F% L, k6 g& {3 H7 p$ TIndicators (KPI) such as delay, jitter or loss at limited cost.
' g g" X' r9 fIn this paper we propose RouteNet, a novel network model based
' O0 @: e' u. h; b! ~% k {on Graph Neural Network (GNN) that is able to understand" |4 m( ?1 |1 K) [
the complex relationship between topology, routing, and input' N/ l0 S9 e( {$ I% C
traffific to produce accurate estimates of the per-source/destination7 O1 d2 D$ w0 o2 O7 |& m4 P5 l
per-packet delay distribution and loss. RouteNet leverages the- g5 A% v! C5 b
ability of GNNs to learn and model graph-structured information; z* r( H& t& W0 i! A5 |7 X
and as a result, our model is able to generalize over arbitrary
6 r+ n) N* q {: ~! z' L$ Xtopologies, routing schemes and traffific intensity. In our eval6 K# s( D5 _" @9 y; B
uation, we show that RouteNet is able to predict accurately8 M: G$ y( `8 w
the delay distribution (mean delay and jitter) and loss even in
& J. Q% S7 G" ~, d) z! @2 gtopologies, routing and traffific unseen in the training (worst case
5 _- E" h! S+ dMRE = 15.4%). Also, we present several use cases where we
* J! v) ?: O( A. oleverage the KPI predictions of our GNN model to achieve
, l- u1 G$ \: e7 v- @; Y1 Meffificient routing optimization and network planning.
7 W9 D% m4 P. \6 X* x& v6 H+ t% H1 r3 e
, P+ A, a( P( G5 O) ~2 `
-
-
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 |