NFV and SDN: The enablers for elastic networks Kashif Mahmood, Next Generation Network Technology (NGNT) Group, Telenor Research, Norway
UNIK Guest Lecture, 8th May 2014
Outline
• Challenges What is coming
• The Network of today
• Elastic Networks Way forward
• Enablers for Elastic networks NFV / SDN
• Use cases – vCRAN – vEPC – vCDN
• Conclusion and Outlook
2
Outline
• Challenges What is coming
• The Network of today
• Elastic Networks Way forward
• Enablers for Elastic networks NFV / SDN
• Use cases – vCRAN – vEPC – vCDN
• Conclusion and Outlook
3
Challenges: Video dominated exponential data growth
Voice to data
4
*Traffic does not include DVB-H, Wi-Fi, or Mobile WiMax. Voice does not include VoIP.
*Source : Ericson mobility report
5 *Traffic does not include DVB-H, Wi-Fi, or Mobile WiMax. Voice does not include VoIP.
*Source : Ericson mobility report Voice to data Exponential data growth
Challenges: Video dominated exponential data growth
6 *Traffic does not include DVB-H, Wi-Fi, or Mobile WiMax. Voice does not include VoIP.
*Source : Ericson mobility report Voice to data Exponential data growth Mobile video traffic 69% of data traffic by the end of 2018 ~CVI Roughly a quarter of Netflix subscribers already streaming video to their smartphones, ~RCR Wireless
Challenges: Video dominated exponential data growth
7
Challenges: Traffic growth
Time
Video dominated exponential data growth Diverse traffic
Challenges: Diverse traffic
Traffic type : Emergency services, Machine to Machine, IoT … Signaling storm “Always on” nature of mobile networks
8
Challenges: Diverse traffic
Traffic type : Emergency services, Machine to Machine, IoT … Signaling storm “Always on” nature of mobile networks
9
10
Challenges: Traffic growth
Time
*Soure: Prof Panagiotis in Expert workshop, Stuttgart
Video dominated exponential data growth Diverse traffic
Ultra dense heterogeneous networks
Challenges: Ultra dense heterogeneous networks
3-10 or more small cells per macro cells cells shrinking (from macro to no cell concept)
11
*Soure: Prof Panagiotis in Expert workshop, Stuttgart
12
Challenges:
Video dominated exponential data growth Diverse traffic
Ultra dense heterogeneous networks
No more competition on the basis of network coverage but on the basis of
features and services.
Traffic growth
Time
*Soure: Prof Panagiotis in Expert workshop, Stuttgart
Outline
• Challenges What is coming
• The Network of today
• Elastic Networks Way forward
• Enablers for Elastic networks NFV / SDN
• Use cases – vCRAN – vEPC – vCDN
• Conclusion and Outlook
13
The Network of today
14
MME
SGW
HSS
PGW
IP network
PCRF
eNB
eNB UE
SGi-LAN
S1-u
S1-c
X2 S11
S5
core Access
The Network of today
15
MME
SGW
HSS
PGW
IP network
PCRF
eNB
eNB UE
SGi-LAN
S1-u
S1-c
X2 S11
S5
core Access
SGi-LAN S5
Radio
Layer :1
Physical layer
Layer : 3
Layer :2
RRC
PDCP RLC MAC
RRM
UE
The Network of today
16
MME
SGW
HSS
PGW
IP network
PCRF
eNB
eNB UE
SGi-LAN
S1-u
S1-c
X2 S11
S5
Core Access
• Controls flow-based charging functionalities in PGW • Ensures compliance with users subscription
The Network of today
17
MME
SGW
HSS
PGW
IP network
PCRF
eNB
eNB UE
SGi-LAN
S1-u
S1-c
X2 S11
S5
core Access
• UL and DL service level charging (input from PCRF, DPI) • IP address allocation for UE • QoS enforcement • Mobility anchor for non-3GPP technologies (CDMA, WiMAX)
The Network of today
18
MME
SGW
HSS
PGW
IP network
PCRF
eNB
eNB UE
SGi-LAN
S1-u
S1-c
X2 S11
S5
core Access
• Local Mobility anchor when UE moves between eNBs • Temporarily buffers data when MME initiates paging of UE • Mobility anchor for GPP technologies (GPRS, UMTS)
The Network of today
19
MME
SGW
HSS
PGW
IP network
PCRF
eNB
eNB UE
SGi-LAN
S1-u
S1-c
X2 S11
S5
core Access
• Processes signaling between UE and core • Coordination and scheduling of resources related to location of UE • Establishment, maintenance and release of the bearers • Functions related to inter-working with other networks e.g. to handle voice calls, CSFB (Circuit switch fall back) • Signaling security
The Network of today
20
MME
SGW
HSS
PGW
IP network
PCRF
eNB
eNB UE
SGi-LAN
S1-u
S1-c
X2 S11
S5
core Access
• Users subscription data such as QoS profile • Information about which PDNs can it connect to • Authentication
The Network of today
21
MME
SGW
HSS
PGW
IP network
PCRF
eNB
eNB UE
SGi-LAN
S1-u
S1-c
X2 S11
S5
core Access
Firewall Video opt. DPI
TCP Opt
UE
UE
core Access
UE UE UE
Characteristic of the Network of today
22
Topologically fixed network nodes
Many redundant functions repeated in each black box
Takes a long tome to roll out new services New services means new equipment
NOT application aware.
Statistics of the network difficult to obtain due to the distributed nature of network
Control logic in each device difficult to upgrade
Designed for maximum load but traffic load is fluctuating 80% of base stations (BS) capacity and upto half of core networks capacity unused,~SoftEPC, ICC’13
policy policy
policy
auth
auth
Characteristic of the Network of today
23
Topologically fixed network nodes
Many redundant functions repeated in each black box
Takes a long tome to roll out new services New services means new equipment
NOT application aware.
Statistics of the network difficult to obtain due to the distributed nature of network
Control logic in each device difficult to upgrade
Designed for maximum load but traffic load is fluctuating 80% of base stations (BS) capacity and upto half of core networks capacity unused,~SoftEPC, ICC’13
Characteristic of the Network of today
24
Topologically fixed network nodes
Many redundant functions repeated in each black box
Takes a long tome to roll out new services New services means new equipment
NOT application aware.
Statistics of the network difficult to obtain due to the distributed nature of network
Control logic in each device difficult to upgrade
Designed for maximum load but traffic load is fluctuating 80% of base stations (BS) capacity and upto half of core networks capacity unused,~SoftEPC, ICC’13
Characteristic of the Network of today
25
Topologically fixed network nodes
Many redundant functions repeated in each black box
Takes a long tome to roll out new services New services means new equipment
NOT application aware.
Statistics of the network difficult to obtain due to the distributed nature of network
Control logic in each device difficult to upgrade
Designed for maximum load but traffic load is fluctuating 80% of base stations (BS) capacity and upto half of core networks capacity unused,~SoftEPC, ICC’13
Characteristic of the Network of today
26
Topologically fixed network nodes
Many redundant functions repeated in each black box
Takes a long tome to roll out new services New services means new equipment
NOT application aware.
Statistics of the network difficult to obtain due to the distributed nature of network
Control logic in each device difficult to upgrade
Designed for maximum load but traffic load is fluctuating 80% of base stations (BS) capacity and upto half of core networks capacity unused,~SoftEPC, ICC’13
Characteristic of the Network of today
27
Topologically fixed network nodes
Many redundant functions repeated in each black box
Takes a long tome to roll out new services New services means new equipment
NOT application aware.
Statistics of the network difficult to obtain due to the distributed nature of network
Control logic in each device difficult to upgrade
Designed for maximum load but traffic load is fluctuating 80% of base stations (BS) capacity and upto half of core networks capacity unused,~SoftEPC, ICC’13
Traffic and revenue
28
Time
Voice Era
Data Era
*Trend from unstrung Pyramid research
Traffic
Traffic and Revenue : Traditional Networks
29
Time
Voice Era
Data Era
*Trend from unstrung Pyramid research
Gap widening
revenue
Traffic
Traffic and Revenue : Traditional Networks
Traffic and Revenue : Traditional Networks
30
Time
revenue
Traffic
Voice Era
Data Era
*Trend from unstrung Pyramid research
Traffic and Revenue : Elastic networks
31
Time
revenue Traffic
Voice Era
Data Era
*Trend from unstrung Pyramid research
Enablers ?
Outline
• Challenges What is coming
• The Network of today
• Elastic Networks Way forward
• Enablers for Elastic networks NFV / SDN
• Use cases – vCRAN – vEPC – vCDN
• Conclusion and Outlook
32
Traditional networks Elastic Networks
33
Traditional network
33
MME
SGW
HSS
PGW
IP
PCRF
eNB
Elastic network
Traditional networks Elastic Networks
34 34
MME
SGW
HSS
PGW
IP
PCRF
eNB
Elastic network
IP
MME
HSS
SGW
PCRF
eNB v.opt tcp.opt
DPI
FW PGW
Traditional network
Can scale up and down as well as move resources
Traditional networks Elastic Networks
35 35
MME
SGW
HSS
PGW
IP
PCRF
eNB
Elastic network
IP
MME
HSS
SGW
PGW FW
PCRF
eNB v.opt tcp.opt
DPI
FW FW PGW
Traditional network
Can scale up and down as well as move resources
Traditional networks Elastic Networks
36 36
MME
SGW
HSS
PGW
IP
PCRF
eNB
Elastic network
IP
MME
HSS
SGW
PGW FW
PCRF
eNB v.opt tcp.opt
DPI
FW FW PGW
Traditional network
Can scale up and down as well as move resources
Enablers: NFV, SDN
Outline
• Challenges What is coming
• The Network of today
• Elastic Networks Way forward
• Enablers for Elastic networks NFV / SDN
• Use cases – vCRAN – vEPC – vCDN
• Conclusion and Outlook
37
Enablers: NFV and SDN
38
Time
revenue Traffic
Voice Era
Data Era
*Trend from unstrung Pyramid research
• NFV • SDN • …
Network function Virtualization (NFV)
• Network functions are implemented on virtual networks VNFs • Almost the same features but more flexibility. • Uses custom of the shelf (COTS) equipment
39
VNF N
…
VNF 1
VNF 2
PCRF
SGW …
MME
Architecture NFV
40
Virtual Network Functions (VNFs)
VNF VNF VNF VNF VNF
Architecture of network virtualization
*source : ETSI
41
Virtual Network Functions (VNFs)
VNF VNF VNF VNF VNF
NFV Infrastructure (NFVI)
Architecture of network virtualization
*source : ETSI
42
Virtual Network Functions (VNFs)
VNF VNF VNF VNF VNF
NFV Infrastructure (NFVI)
Virtual Compute
Virtual Storage
Virtual Network
Architecture of network virtualization
*source : ETSI
43
Virtual Network Functions (VNFs)
VNF VNF VNF VNF VNF
NFV Infrastructure (NFVI)
Compute Storage Network
Virtual Compute
Virtual Storage
Virtual Network
Hardware resources
Architecture of network virtualization
*source : ETSI
44
Virtual Network Functions (VNFs)
VNF VNF VNF VNF VNF
NFV Infrastructure (NFVI)
Compute Storage Network
Virtual Compute
Virtual Storage
Virtual Network
Virtualization Layer
Hardware resources
Architecture of network virtualization
*source : ETSI
45
Virtual Network Functions (VNFs)
VNF VNF VNF VNF VNF
NFV Management
and Orchestration
NFV Infrastructure (NFVI)
Compute Storage Network
Virtual Compute
Virtual Storage
Virtual Network
Virtualization Layer
Hardware resources
Architecture of network virtualization
46
Virtual Network Functions (VNFs)
VNF VNF VNF VNF VNF
NFV Management
and Orchestration
NFV Infrastructure (NFVI)
Compute Storage Network
Virtual Compute
Virtual Storage
Virtual Network
Virtualization Layer
Hardware resources
VNF
VIM/NW controller
VNF
VNF
Network service
Customer self service portal
47
Virtual Network Functions (VNFs)
VNF VNF VNF VNF VNF
NFV Management
and Orchestration
NFV Infrastructure (NFVI)
Compute Storage Network
Virtual Compute
Virtual Storage
Virtual Network
Virtualization Layer
Hardware resources
VNF
VNF
VNF
Network service
Customer self service portal
VIM Network controller
One of the ways to control the network ?
48
Virtual Network Functions (VNFs)
VNF VNF VNF VNF VNF
NFV Management
and Orchestration
NFV Infrastructure (NFVI)
Compute Storage Network
Virtual Compute
Virtual Storage
Virtual Network
Virtualization Layer
Hardware resources
VNF
VNF
VNF
Network service
Customer self service portal
SDN : One of the ways to control the network
VIM SDN controller
Enabler: Software Defined Networking (SDN)
49
North bound interface (REST)
South bound interface (OpenFlow)
API API API
SDN Controller
Separate the data plane and the control plane
Data plane
SDN at work !
• Speaking at the Open Networking Summit in April 2012, Urs Holzle, senior vice-president of technical infrastructure at Google said
• “The company can now prioritize certain traffic, such as Gmail backups, to ensure they get through in a timely manner”
• Perhaps most impressively, Holzle told the conference that, “in using SDN to intelligently manage the flow of traffic through its internal network, Google will eventually hit 100% network utilization. In an industry where 30% to 40% is considered standard, that is a huge performance increase.”
50
Outline
• Challenges What is coming
• The Network of today
• Elastic Networks Way forward
• Enablers for Elastic networks NFV / SDN
• Use cases – vCRAN – vEPC – vCDN
• Conclusion and Outlook
51
vCRAN
52
micro
pico
pico micro
macro macro
x2
EPC
BBU Base Band Unit RRH Remote Radio Head CPRI Common Public Radio Interface
VNF 1
VNF N
VNF 1
VNF N
BBU pool
VNF 1
VNF N
VNF 1
VNF N
BBU pool
Internet
CPRI*
RRH
RRH
RRH
Backhaul
Fronthaul CWDM, Microwave (mm)
vCRAN Why • A generic network, supporting different tech 3G, WiFi, 4G, XG and different applications • Allocating spectrum resources dynamically
53
vCRAN Why • A generic network, supporting different tech 3G, WiFi, 4G, XG and different applications • Allocating spectrum resources dynamically
• LTE-A will use
– network MIMO – coordinated scheduling – Interference management
54
CRAN
vCRAN Why • A generic network, supporting different tech 3G, WiFi, 4G, XG and different applications • Allocating spectrum resources dynamically
• LTE-A will use
– network MIMO – coordinated scheduling – Interference management
• Liberty of using the slice specific to the type of application and technology
55
CRAN
vCRAN Why • A generic network, supporting different tech 3G, WiFi, 4G, XG and different applications • Allocating spectrum resources dynamically
• LTE-A will use
– network MIMO – coordinated scheduling – Interference management
• Liberty of using the slice specific to the type of application and technology
• RAN sharing – Active Only cell sites – Passive In addition to cell sites radio spectrum and baseband processing unit is shared
56
CRAN
vCRAN Why • A generic network, supporting different tech 3G, WiFi, 4G, XG and different applications • Allocating spectrum resources dynamically
• LTE-A will use
– network MIMO – coordinated scheduling – Interference management
• Liberty of using the slice specific to the type of application and technology
• RAN sharing – Active Only cell sites – Passive In addition to cell sites radio spectrum and baseband processing unit is shared
• Age old algorithms on user mobility and traffic patterns can be put into practice – leveraging global view and efficient monitoring tools of SDN
• CRAN Green RAN
57
CRAN
vCRAN, Challenges
58
• For centralized CRAN architecture
• High BW requirement to carry baseband I/Q between BBU and RRH
• Fairness in power, spectrum or product of two
• Fully centralized architecture: requires high capacity fronthaul
• Integration with backhaul.
• CPRI CWDM, Microwave ?
• Coordination
• Low latency and near zero jitter
• Virtualized solution vs DSP
Fronthaul
Backhaul
Outline
• Challenges What is coming
• The Network of today
• Elastic Networks Way forward
• Enablers for Elastic networks NFV / SDN
• Use cases – vCRAN – vEPC – vCDN
• Conclusion and Outlook
59
vEPC
60
vCDN
61
core
backhaul
VoIP
Internet GW
GW
GW
CDN
Access
vCDN
62
core
backhaul
VoIP
Internet
Access
GW
GW
GW
CDN
CDN
In a Nutshell • Reduction in CAPEX and OPEX.
– Cost effective Commodity hardware a plentiful and scalable resource.
• Address new requirements through new software not new equipment. • Fast service delivery • Bandwidth on demand services
• Integrated RAN Each slice for different traffic type and technology. Selected by SDN controller. The slice can be scaled up/down NFV
• Resource pool in the cloud on VMs NFV
New abstraction view of networks possible ~ courtesy SDN Challenge: Cost and Performance analysis of virtualizing network functions
63
In a Nutshell • Reduction in CAPEX and OPEX.
– Cost effective Commodity hardware a plentiful and scalable resource.
• Address new requirements through new software not new equipment. • Fast service delivery • Bandwidth on demand services
• Integrated RAN Each slice for different traffic type and technology. Selected by SDN controller. The slice can be scaled up/down NFV
• Resource pool in the cloud on VMs NFV
New abstraction view of networks possible ~ courtesy SDN Challenge: Cost and Performance analysis of virtualizing network functions
64
Cost and Performance Analysis : Expected Outcome
Cost Benefit
Performance
MME
CDN
DPI IMS
MGW
PCRF
Voice Service
Contributors
Low level variable Description
Strategic Results variable
Link utilization Percentage of total link utilization between two points in the network. VNFs can be located at different locations in the network Impacts link utilization.
Cost, performance
Resource utilization
NFV has the potential to reduce resource utilization since multiple VNFs can be installed on COTS hardware. How will the virtualisation of the function impact utilization of the hardware, i.e. router, switches, servers and storage?
Cost, performance
Energy consumption
Can be reduced by consolidating equipment and exploit power management features in standard servers and storage, as well as workload consolidation and location optimization.
Cost, performance
Performance Analysis
66
Low level variable Description
Strategic Results variable
Link utilization Percentage of total link utilization between two points in the network. VNFs can be located at different locations in the network Impacts link utilization.
Cost, performance
Resource utilization
NFV has the potential to reduce resource utilization since multiple VNFs can be installed on COTS hardware. How will the virtualisation of the function impact utilization of the hardware, i.e. router, switches, servers and storage?
Cost, performance
Energy consumption
Can be reduced by consolidating equipment and exploit power management features in standard servers and storage, as well as workload consolidation and location optimization.
Cost, performance
Portability Concerns the capabilities to load, execute and move software functions across different but standard data centres and network locations.
Cost
Elasticity Concerns the capabilities to provide an easier way to scale up/down and in/out hardware and software resources as traffic demands increase/decreases.
Cost
67
Performance Analysis
Low level variable Description
Strategic Results variable
Link utilization Percentage of total link utilization between two points in the network. VNFs can be located at different locations in the network Impacts link utilization.
Cost, performance
Resource utilization
NFV has the potential to reduce resource utilization since multiple VNFs can be installed on COTS hardware. How will the virtualisation of the function impact utilization of the hardware, i.e. router, switches, servers and storage?
Cost, performance
Energy consumption
Can be reduced by consolidating equipment and exploit power management features in standard servers and storage, as well as workload consolidation and location optimization.
Cost, performance
Portability Concerns the capabilities to load, execute and move software functions across different but standard data centres and network locations.
Cost
Elasticity Concerns the capabilities to provide an easier way to scale up/down and in/out hardware and software resources as traffic demands increase/decreases.
Cost
Realtime properties
Delay, jitter, latency etc. Performance
Security How will the transition from physical to virtual network functions impact security? The role of SDN as a centralized controller could have an effect on security
Performance
68
Performance Analysis
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑀𝑀𝑀 = 𝑤𝑀𝑀𝑀𝐶𝐵,𝐿𝐿 . 𝐿𝐶𝐶𝐿 𝑈𝐶𝐶𝑈𝐶𝑈𝑈𝐶𝐶𝐶𝐶 + 𝑤𝑀𝑀𝑀
𝐶𝐵,𝑃𝑃𝑃.𝑃𝐶𝑃𝐶𝑈𝑃𝐶𝑈𝐶𝐶𝑃 + ⋯ 𝑃𝐶𝑃𝐶𝐶𝑃𝑃𝑈𝐶𝑃𝐶𝑀𝑀𝑀 = 𝑤𝑀𝑀𝑀
𝑃𝑃𝑃,𝐿𝐿 .𝐿𝐶𝐶𝐿 𝑈𝐶𝐶𝑈𝐶𝑈𝑈𝐶𝐶𝐶𝐶 + 𝑤𝑀𝑀𝑀𝑃𝑃𝑃,𝑀𝐶 .𝐸𝐶𝐶𝑃𝐸𝑃𝐶𝐶𝐶𝐶𝐸𝑃𝐸𝐶𝐶𝐶𝐶 + ⋯
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑃𝐶𝑃𝑃 = 𝑤𝑃𝐶𝑃𝑃
𝐶𝐵,𝐿𝐿 . 𝐿𝐶𝐶𝐿 𝑈𝐶𝐶𝑈𝐶𝑈𝑈𝐶𝐶𝐶𝐶 + 𝑤𝑃𝐶𝑃𝑃𝐶𝐵,𝑃𝑃𝑃.𝑃𝐶𝑃𝐶𝑈𝑃𝐶𝑈𝐶𝐶𝑃 + ⋯
𝑃𝐶𝑃𝐶𝐶𝑃𝑃𝑈𝐶𝑃𝐶𝑃𝐶𝑃𝑃 = 𝑤𝑃𝐶𝑃𝑃𝑃𝑃𝑃,𝐿𝐿 .𝐿𝐶𝐶𝐿 𝑈𝐶𝐶𝑈𝐶𝑈𝑈𝐶𝐶𝐶𝐶 + 𝑤𝑃𝐶𝑃𝑃
𝑃𝑃𝑃,𝑀𝐶 .𝐸𝐶𝐶𝑃𝐸𝑃𝐶𝐶𝐶𝐶𝐸𝑃𝐸𝐶𝐶𝐶𝐶 + ⋯
⋮
Performance Analysis : Expected Outcome
Cost Benefit
Performance
MME
CDN
DPI IMS
MGW
PCRF
Voice Service
Final-line • Challenges What is coming
• The Network of today
• Elastic Networks Way forward
• Enablers for Elastic networks NFV / SDN
• Use cases
– vCRAN – vEPC – vCDN
• Conclusion
70