Env2Vec Accelerating VNF Testing with Deep Learning Guangyuan Piao, Pat Nicholson, Diego Lugones Eurosys 2020
Release Operate Deploy Build Code VNF Developer Customer Network
Release Operate Deploy Build Code New features VNF Developer Customer Network
Upgrades Release Operate Deploy Build Code New features VNF VNF Developer testing ü 5 9’s ü SLO’s Customer ü Assurance Network ü … time-consuming
VNF testing VNF ? Middleware Config OS KPI’s Virt/Container HW Cloud stack
VNF Anomaly ! testing VNF ? Middleware Config OS Virt/Container HW KPI’s Cloud stack ML
VNF VNF Different stack Middleware Middleware Config Config Cloud stack OS OS Virt/Container Virt/Container HW HW high-dimensional parameter space
Env2Vec Accelerating VNF Testing with Deep Learning 1) Robust to environment variations 2) Simple single ML model 3) Work in previously unseen environments VNF Middleware Config OS Virt/Container HW VNF Middleware Config OS Virt/Container HW
Env2Vec Accelerating VNF Testing with Deep Learning Deep Learning Architecture Anomaly detection FNN GRU Embeddings Contextual Historical Environment FNN: FeedForward Neural Network Features data Metadata GRU: Gated Recurrent Units VNF Middleware Config … g OS n i n i a r T Virt/Container HW
Book (wikipedia) embeddings by genre Environments Embeddings by test case @github: WillKoehrsen/wikipedia-data-science Build type D (debug), T (test), S (stable), etc.
Env2Vec Accelerating VNF Testing with Deep Learning Deep Learning Architecture Anomaly ! KPI’s VNF timeseries Anomaly detection Middleware Config OS Virt/Container FNN GRU Embeddings HW e m i t n u R Contextual Historical Environment Features data Metadata VNF Middleware Config … g OS n i n i a r T Virt/Container HW
KDN dataset (public) Open Virtual Switch Snort SDN-enabled firewall Carrier-grade VNF for multiple Testing environments Build types Services Under test Accuracy 86.2% - 100% False alarms reduced by 20.9% to 38.1% Simplified adoption: single model competitive against multi-model proposals
Env2Vec Accelerating VNF Testing with Deep Learning Guangyuan Piao, Pat Nicholson, Diego Lugones Eurosys 2020
Recommend
More recommend