5G Networks Slicing and Function Placement

5G Networks: Slicing and Function Placement
Wolfgang Kellerer
This work is part of a project that has received funding from the European Research Council (ERC)
under the European Union’s Horizon 2020 research and innovation program grant agreement No 647158
Technical University of Munich
Project ID 16KIS0473 – FlexNets (2015 – 2020).
5G Networks – my viewpoint
Security
IT Cloud Edge
Data Center Data Centers
Network design Compact &
distributed
Resilience
Performance Real‐time
capable
Scalability
Telco optimized Flexibility Throughput
optimized
SDN / NFV SDN / NFV
Cloud Mgmt. System Cloud Mgmt. System
SDN Network Slicing
Distributed Cloud Orchestration
NFV
Project ID 16KIS0473
Slicing and Network Functions
Flexibility [1] to react to dynamic requests
for
• Slicing: embedding and isolation

• Network functions placement [2]
Network softwarization helps:

SDN, NV, NFV as enablers
Source: NGMN 5G white paper
[1] W. Kellerer, et al: How to Measure Network Flexibility? A Proposal for Evaluating Softwarized Networks. IEEE Communications Magazine, 2018
[2] A. Basta, W. Kellerer, M. Hoffmann, H. Morper, K. Hoffmann, Applying NFV and SDN to LTE Mobile Core Gateways; The Functions Placement Problem,

ACM SICGOMM 2014 Workshop on AllThingsCellular (ATC), Chicago, IL, USA, August 2014.
Network Softwarization can provide a solution
Unleashing 5G flexibility through combining NV + SDN + NFV

Possible Realization: HyperFLEX [3]
bring your own SDN controller • HyperFLEX: SDN virtualization layer
for flexible, reliable and dynamic SDN slicing
and interconnecting Virtual Network Functions
control plane
SDN Hypervisor • On-demand virtual network resources

abstraction translation isolation
to the service providers, i.e. tenants
• Dynamic changes during the run-time –

coping with VNF migrations
• Isolation and performance guarantees for the tenant
• Multiple tenants or service providers coexist

data plane on the same network without performance degradation
[3] A. Blenk, A. Basta, W. Kellerer. HyperFlex: An SDN virtualization architecture with flexible hypervisor function allocation.

Integrated Network Management (IM), 2015 IFIP/IEEE International Symposium on. IEEE, 2015.
HyperFLEX Architecture
Latency, CPU, … SDN Slice
Tenant View HyperFlex View Performance Monitoring Tenant SDN Controller 3 different views
Control-plane
Isolation
Tenant View HyperFlex Performance SDN Control isolation in

Controller View Controller Monitoring DB Network
admission control software
checks available HyperFlex Core or hardware
resources (CPU,...) Admission CTR NET
Control Controller SDN Hypervisors
Dynamic Migration
interfaces to
Slice
State DB
Embedding FlowVisor OVX DITRA ….. different
slice embedding hypervisors
(virtual network)
Virtual Link Virtual Node
SDN Data
Network Data-plane
Isolation
OF Switch
HyperFLEX: Admission Control
• Automated request of virtual SDN slices

• Guarantees for control plane, i.e., mapping of slice requests to HV resources
• Run time update to slice
• Embedding of virtual links on the physical network
(a) Tenant View (b) HyperFlex View
HyperFLEX: Isolation
• correlation between control message rate and hypervisor CPU consumption
• assign control plane rate to each slice → hypervisor CPU isolation
• change isolation option on run time: “net“ vs. “soft“
• vSDN2 exceeds its SLA

• isolation minimizes impact on vSDN1 latency
isolation options
• net: no control plane loss, at cost of latency

• soft: low latency, at cost of possible loss
HyperFLEX in Operation
How can we boost the solving of the related optimization problems?
Problem Optimization produce

Instances Algorithm Problem
Solutions
Traditional System
learn from (offline)
Problem Machine Solution Optimization produce

Instances Learning Information Algorithm Problem
Solutions
o’zapft is framework [4]
• State-of-the-art: Neglects produced data!

• Idea: Use problem/solution data generated by algorithms regularly solving problems
[4] A. Blenk, P. Kalmbach, S. Schmid, W. Kellerer: o'zapft is: Tap Your Network Algorithm's Big Data!
ACM SIGCOMM 2017 Wrksp. on Big Data Analytics and Machine Learning for Data Communication Networks (Big‐DAMA), 2017.
Data Available: P. Kalmbach, J. Zerwas, M. Manhart, A. Blenk, S. Schmid, W. Kellerer. Data on "o’zapft is Tap Your Network Algorithm’s Big Data!”,2017 https://doi.org/10.14459/2017md136158910
Case Study: Predicting Acceptance Probabilities of VNE Requests
• Supervised learning: use data with accepted and rejected requests! Offline training!
• Recurrent neural network (RNN) for classification
• Filter infeasible and requests with unacceptable algorithm runtime (“no solution“)
11
Can we speed-up optimal algorithms using admission control?
• Efficient Filtering of infeasible and unacceptable requests

• Efficient saving of model creation time
12
Latest Results: Neurovine [5]
Hopfield neural network to preprocess (subgraph extraction) VNE algorithms
 tailored filtering
• Idea: Extract subgraph with physical nodes close to each other and high
available capacities
[5] A. Blenk, P. Kalmbach, J. Zerwas, M. Jarschel, S. Schmid, W. Kellerer: NeuroViNE: A Neural Preprocessor for Your Virtual Network Embedding Algorithm
IEEE INFOCOM 2018 (main conference), Honolulu, HI, USA, April 15‐19, 2018.
Neurovine: Efficiency on Real Network Topologies
• VNE algorithms (GRC, DViNE, RViNE) vs. Hopfield variants (HF-GRC, HF-DViNE, HF-RViNE)
• NeuroViNE accepts more networks with less costs
14
Conclusion
Key takeaways
• 5G  flexibility for slicing and function placement

combining network softwarization enablers: SDN + NV + NFV
• Example for effective combination keeping the flexibility: HyperFLEX
• Speedup slice embedding and function placement through Machine

Learning-based preprocessing!
Key References
Andreas Blenk, Patrick Kalmbach, Johannes Zerwas, Michael Jarschel, Stefan Schmid, Wolfgang Kellerer:
NeuroViNE: A Neural Preprocessor for Your Virtual Network Embedding Algorithm
IEEE INFOCOM 2018 (main conference), Honolulu, HI, USA, April 15‐19, 2018.
A. Blenk, P. Kalmbach, S. Schmid, W. Kellerer: o'zapft is: Tap Your Network Algorithm's Big Data!

ACM SIGCOMM 2017 Wrks. on Big Data Analytics and Machine Learning for Data Communication Networks (Big‐DAMA), 2017.
A. Basta, W. Kellerer, M. Hoffmann, H. Morper, K. Hoffmann,
Applying NFV and SDN to LTE Mobile Core Gateways; The Functions Placement Problem,
ACM SICGOMM 2014 Workshop on AllThingsCellular (ATC), Chicago, IL, USA, August 2014.
A. Blenk, A. Basta, J. Zerwas, M. Reisslein, W. Kellerer,
Control Plane Latency with SDN Network Hypervisors: Cost of Virtualization,
IEEE Transactions on Network and Service Management, Sept. 2016.
A. Blenk, A. Basta, W. Kellerer.
HyperFlex: An SDN virtualization architecture with flexible hypervisor function allocation.
Integrated Network Management (IM), 2015 IFIP/IEEE International Symposium on. IEEE, 2015.
HyperFLEX: https://github.com/tum‐lkn/HyperFLEX
W. Kellerer, et al: How to Measure Network Flexibility? A Proposal for Evaluating Softwarized Networks.

IEEE Communications Magazine, 2018.

5G Networks Slicing and Function Placement

Uploaded by

Copyright:

Available Formats

5G Networks Slicing and Function Placement

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

5G Networks Slicing and Function Placement

Uploaded by

Copyright:

Available Formats

5G Networks: Slicing and Function Placement

Flexibility [1] to react to dynamic requests

• Slicing: embedding and isolation

Network softwarization helps:

[1] W. Kellerer, et al: How to Measure Network Flexibility? A Proposal for Evaluating Softwarized Networks. IEEE Communications Magazine, 2018

[2] A. Basta, W. Kellerer, M. Hoffmann, H. Morper, K. Hoffmann, Applying NFV and SDN to LTE Mobile Core Gateways; The Functions Placement Problem,

Unleashing 5G flexibility through combining NV + SDN + NFV

SDN Hypervisor • On-demand virtual network resources

• Dynamic changes during the run-time –

• Isolation and performance guarantees for the tenant

• Multiple tenants or service providers coexist

[3] A. Blenk, A. Basta, W. Kellerer. HyperFlex: An SDN virtualization architecture with flexible hypervisor function allocation.

Latency, CPU, … SDN Slice

Tenant View HyperFlex Performance SDN Control isolation in

• Automated request of virtual SDN slices

• vSDN2 exceeds its SLA

• net: no control plane loss, at cost of latency

Problem Optimization produce

Problem Machine Solution Optimization produce

• State-of-the-art: Neglects produced data!

• Efficient Filtering of infeasible and unacceptable requests

• 5G  flexibility for slicing and function placement

• Example for effective combination keeping the flexibility: HyperFLEX

• Speedup slice embedding and function placement through Machine

A. Blenk, P. Kalmbach, S. Schmid, W. Kellerer: o'zapft is: Tap Your Network Algorithm's Big Data!

W. Kellerer, et al: How to Measure Network Flexibility? A Proposal for Evaluating Softwarized Networks.

You might also like