0% found this document useful (0 votes)
76 views

On Service Chaining Using Virtual Network Functions in Operator Networks

This document summarizes a research paper on optimizing service chain deployment using virtual network functions in operator networks. The paper formulates the problem of jointly placing VNFs and routing traffic as an integer linear program to minimize network resource consumption. It analyzes different service chaining strategies across varying CPU core counts and traffic loads. The paper finds that determining the "inflection point" for core count and traffic can achieve near-optimal resource usage, reducing it by almost 50%. Congestion and load balancing aspects are also discussed to infer tradeoffs between centralized and distributed deployment strategies.

Uploaded by

claus8891
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
76 views

On Service Chaining Using Virtual Network Functions in Operator Networks

This document summarizes a research paper on optimizing service chain deployment using virtual network functions in operator networks. The paper formulates the problem of jointly placing VNFs and routing traffic as an integer linear program to minimize network resource consumption. It analyzes different service chaining strategies across varying CPU core counts and traffic loads. The paper finds that determining the "inflection point" for core count and traffic can achieve near-optimal resource usage, reducing it by almost 50%. Congestion and load balancing aspects are also discussed to infer tradeoffs between centralized and distributed deployment strategies.

Uploaded by

claus8891
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 43

On Service Chaining using

Virtual Network Functions in


Operator Networks

By
Abhishek Gupta
Department of Computer Science
University of California, Davis

July 29, 2016

Acknowledgement: NSF Grant No. CNS-1217978


Network Function Virtualization (NFV)

9/7/20 2
Service Chain

9/7/20 [1] Deutsche Telekom, “Flexible Service Chaining : Requirements and Architecture”, 2014 3
Network-enabled Cloud

9/7/20 [2] Ericsson, “The real-time cloud combining cloud, NFV and service provider SDN,” 2014 4
Service Chain Deployment (VNF Placement and
Routing)

ISI Delhi
Encryption service chain Virtualization resource

VNF Placement
ISI Kolkata
Routing

Operator Network
(BSNL, Airtel…)
ISI Bengaluru
5
Problem Description

• Given
• Network topology
• Capacity of link
• Set of DC locations
• Set of NFV-capable nodes (in addition to the DC)
• Traffic flows between source-destination pairs
• Set of k-shortest paths between source-destination pairs
• Set of required network functions (for the service chain)
• The service chain to be deployed
• Number of CPU cores present per NFV-capable node

• Objective
• Minimize the bandwidth consumption in the network by optimal
placement of the VNFs.

[4]A. Gupta et al., “Joint Virtual Network Function Placement and Routing of Traffic in
9/7/20 Operator Networks,” Technical Report, UC Davis
6
Continued…

• Constraints
• Single-path routing
• Capacity constraint for a link (bandwidth)
• Capacity constraint for a node (CPU cores)
• VNF sequence in the service chain (across nodes)
• VNF sequence in the service chain (inside a node)

[4]A. Gupta et al., “Joint Virtual Network Function Placement and Routing of Traffic in
9/7/20 Operator Networks,” Technical Report, UC Davis 7
Integer Linear Program

Objective: Minimize Network


Resource Consumption

Single-path routing constraint

Capacity constraint for link

Capacity constraint for node

VNF sequence in the service chain

VNF sequence in the service chain


(inside a node)

9/7/20 [4]A. Gupta et al., “Joint Virtual Network Function Placement and
Routing of Traffic in Operator Networks,” Technical Report, UC Davis
8
CPU-core-to-throughput relationship of a VNF

Throughput
Applications
1 Gbps 5 Gbps 10 Gbps
NAT 1 CPU 1 CPU 2 CPUs
IPsec VPN 1 CPU 2 CPUs 4 CPUs
Traffic Shaper 1 CPU 8 CPUs 16 CPUs

9/7/20 [3]Cisco, “Cisco Cloud Services Router 1000V 3.14 Series Data Sheet,”2015. 9
VNF placement (on throughput and CPU cores)

s d

f1 f2 f3

s d

NFV-capable nodes

[4]A. Gupta et al., “Joint Virtual Network Function Placement and Routing of Traffic in
9/7/20 Operator Networks,” Technical Report, UC Davis
10
Service Chaining Strategies

• MB only – Middle box (MB) used for service chaining

• DC only (Centralized) – Data center (DC) used for service


chaining

• DC NFV x (Best-Case scenario)– Data center (DC) and ‘x’


NFV-capable nodes used for service chaining. `DC NFV
ALL’ refers to situation where all network nodes are NFV-
capable.

• ALL NFV (Completely Distributed) – A completely


distributed strategy where all nodes are NFV-capable and
there is no DC.

9/7/20 11
Simulation details

9/7/20 12
Continued…

Traffic Flows

9/7/20 13
1 Gbps traffic

9/7/20 14
Continued…

Inflection point

9/7/20 15
Continued…

Inflection
point

9/7/20 16
Continued…

9/7/20 17
Inflection Point (DC NFV ALL)

1 Gbps

18
2.5 Gbps traffic
1 Gbps

9/7/20 [
19
Inflection Point (DC NFV ALL)

2.5 Gbps

20
Results

1 Gbps 2.5 Gbps

7.5 Gbps 5 Gbps

10 Gbps

9/7/20 21
Inflection points (DC NFV ALL)

9/7/20 22
ALL NFV (Completely-Distributed) vs DC NFV ALL

9/7/20 23
Continued…

ALL NFV infeasible for these CPU core counts at 2.5 Gbps

9/7/20 24
Continued…

9/7/20 25
Continued…

9/7/20 26
Summary

• We formulated an optimization problem for the VNF


placement and routing problem for service chain
deployment by network operators.

• We defined different service chaining strategies and


analyzed the network resource consumption across these
strategies for different CPU core counts and traffic bales.

• We found that by determining the “inflection point” for core


count and offered traffic, we can achieve close-to-optimal
network resource consumption and reduce it by close to
50%.

27
Inferences

• Congestion aspect : In ‘DC–only’ strategy, all flows have to


routed through a single node which will lead to congestion
at DC nodes.
• Congestion at DC nodes can be shown through the infeasibility of
certain nodes to be DC in the ‘DC-only’ scenario in our ILP.
• This infeasibility will occur at different traffic loads for different
nodes in the network.

• Load-balancing (Resource-contention?) aspect : NFV-


capable nodes help reduce the congestion problem on the
single DC node and reduce the operating expenditure for
the network operator.
• Change in network resource between ‘DC-only’, ‘DC NFV 4’ and ‘DC
NFV ALL’ for low, high and infeasible traffic loads for a particular
traffic matrix

28
Network Topology (Internet2)

[5] https://www.internet2.edu/media/medialibrary/2015/08/04/NetworkMap_all.pdf 29
Network Topology (Geant)

[6]
http://geant3plus.archive.geant.net/Resources/Media_Library/PublishingImages/maps/ 30
10G_fibre_2009.jpg
Congestion aspect

• Congestion point calculation


• Nodal degree * capacity < total flows in the network
• Number of paths that include that node for each sd-pair that satisfy
the capacity constraints
• Some anomalies can be explained when the traffic flows terminate at a
certain source or destination.
NSF 14
For each of the Internet2, Geant
Traffic Nodes
Traffic Infeasible DC Nodes
1 Gbps None TM1 TM2 TM3
2.5 Gbps 1,2 1 Gbps None
5 Gbps 4,5,6,7 2.5 Gbps 1,2
7.5 Gbps 11,12,13 5 Gbps 4,5,6,7
,14 7.5 Gbps 11,12,13,14
10 Gbps 3,8,10 10 Gbps 3,8,10
31
Continued..

32
Traffic Matrix? (TMx)

• 3 traffic matrices for each topology.


• Each traffic matrix will have 3 traffic loads.
• Low, high and infeasible loads.
• Infeasible load : traffic load at which ‘DC-only’ strategy is infeasible
for all nodes in the network topology.

33
TM’s for Internet 2

TM1

TM2

TM3

34
TM’s for Geant

TM1

TM2

TM3
35
Load balancing aspect of an NeC

36
Memory is more important than CPU core count

• Will be run on NSF 14 itself


• Memory requirement only there for initial installation
(Inelastic)
• Memory requirement scales with increasing traffic (elastic)
• How to map the memory requirements of each of the
VNF’s?
• Make it uniform like 1 GB for each or map mimicking the
CPU core assigned to each VNF

37
Continued…

38
Thank You

9/7/20 39
Related Work

• [7]S. Mehraghdam et al., “Specifying and Placing Chains of


Virtual Network Functions,”arXiv preprint arXiv:1406.1058, 2014

• [8]A. Mohammadkhan et al., “Virtual Function Placement and


Traffic Steering in Flexible and Dynamic Software Defined
Networks,” IEEE Workshop on Local and Metropolitan Area Networks (LANMAN
15), April 2015

• [9]B. Addis et al., “Virtual Network Functions Placement and


Routing Optimization,” https://hal.inria.fr/hal-01170042/, 2015

• [10]Md. Faizul Bari et al., “On Orchestrating Virtual Network


Functions in NFV,” Computing Research Repository, vol. abs/1503.06377,
2015

9/7/20 40
Ongoing Work

• A column-generation based optimization model for the


current problem (in collaboration with Prof. Brigitte Jaumard, Concordia
University, Montreal)
• Model will scale to provide placement of VNFs for multiple service
chains
• More problem context like latency of VNFs, maximum number of VNF
instances that can be deployed etc. will also be included

• In the problem definition stage for the next problem,


“Towards a service-oriented virtual evolved packet core”,
joint placement of VNFs for EPC functionalities and service chains deployed in
the SGi-LAN (Service Gateway interface LAN)

9/7/20 41
Open Research Problems

• Service chain composition (formalize a request for chaining VNFs


together)

• Deployment of VNF service chains in Multi-Domain


environments

• Virtualization of the mobile core


• EPC
• SGi-LAN
• IMS for VoLTE and video

[10]W. John et al., “Research Directions in Network Service Chaining,” 2013 IEEE SDN
9/7/20 for Future Networks and Services (SDN4FNS)
42
References

• [1] IETF, “Network service chaining problem statement,” 2013.


• [2] Ericsson, “The real-time cloud combining cloud, NFV and service provider SDN,” 2014
• [3] Cisco, “Cisco Cloud Services Router 1000V 3.14 Series Data Sheet,” 2015.
• [4] A. Gupta et al., “Joint Virtual Network Function Placement and Routing of Traffic in Operator
Networks,” Technical Report, UC Davis
• [5] https://www.internet2.edu/media/medialibrary/2015/08/04/NetworkMap_all.pdf
• [6]
http://geant3plus.archive.geant.net/Resources/Media_Library/PublishingImages/maps/10G_fibre_
2009.jpg
• [7] S. Mehraghdam et al., “Specifying and Placing Chains of Virtual Network Functions,”arXiv preprint
arXiv:1406.1058, 2014
• [8] A. Mohammadkhan et al., “Virtual Function Placement and Traffic Steering in Flexible and
Dynamic Software Defined Networks,” IEEE Workshop on Local and Metropolitan Area Networks (LANMAN 15),
April 2015
• [9] B. Addis et al., “Virtual Network Functions Placement and Routing Optimization,”
https://hal.inria.fr/hal-01170042/, 2015
• [10] Md. Faizul Bari et al., “On Orchestrating Virtual Network Functions in NFV,” Computing Research
Repository, vol. abs/1503.06377, 2015
• [11] W. John et al., “Research Directions in Network Service Chaining,” 2013 IEEE SDN for Future Networks
and Services (SDN4FNS)

9/7/20 43

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy