software network opportunities for 5g and · software network opportunities for 5g and ......
TRANSCRIPT
Marius Corici
Fraunhofer FOKUS Institute
Internet: www.open5Gcore.net / www.5G-playground.org
Contact: [email protected]
© M
att
hia
s H
eyd
e /
Fra
un
ho
fer
FO
KU
SSOFTWARE NETWORK OPPORTUNITIES FOR 5G AND
EDGE NETWORKS
2
The biggest change in networking paradigm
Network functions are implemented as software on top
of common hardware
Network can be programmed
NFV/SDN platform acts as an end-to-end middleware
between:
A distributed heterogeneous infrastructure for
compute and storage
Interconnected through a controlled network
Generic network functions implemented in software
running in isolated containers/virtual machines
VPNs, NATs, DNSs, IMSs, EPCs, Application
Servers, etc.
The main value added differentiator between
different solutions is the quality of the software
how well it can solve the specific service needs
Software networks
DC
DC
NF
NF
NF NF NF
NF NF
VNF VNF
VNF
NF
NF
3
A distributed heterogeneous infrastructure including:
Physical components (e.g. radio heads)
heterogeneous data centers (compute & storage)
inter-connecting networks (fronthaul, backhaul, third party backbone, etc.)
Generic network functions implemented in software and running in virtual machines
(e.g. vEPC, vVPN, vNAT, vIMS, …)
Middleware functionality:
Understand the service requirements and transform them into runtime parameters
Brokering the common resources (compute, storage, networking) between multiple
services
Ensuring the end-to-end SLA in dynamic load and network conditions
© Fraunhofer FOKUS
NFV/SDN middleware
4© Fraunhofer FOKUS
NGMN Use cases
© NGMN Alliance https://www.ngmn.org/uploads/media/NGMN_5G_White_Paper_V1_0.pdf
5
5G represents the next generation of network communication services and platforms with initial deployments in 2020
Usage scenarios for IMT2020 and beyond © ITU-T
The 5G Use Cases
Widening of currentcommunication use cases
Low Cost connectivity for hugenumber of devices
Network Islands of Gigabit/s communication
Critical & low latencycommunication
50Mbps anywhereFlexible Networks
6
Flexible compute, storage and memory containers offered to the slice specific software
A set of functions to manage the containers (i.e. Orchestrator)
A set of network functions which execute the specific service (VNFs)
A set of network functions to interact with other slices / end devices
A set of functions to manage the service (i.e. slice management)
A business logic on which all come together
© Fraunhofer FOKUS
Decomposing a slice
7
The software network architecture offers new degrees of flexibility
Customized slices – depending on the specific needs of the devices
With flexible locations – placed at the edge or centralized
With flexible backhaul – intermittent, large delay, etc.
The end to end service has to be offered to the subscribers at the expected levels
Specifying the complete requirements a-priory for a slice requires a complex design,
especially when all the runtime rules have to be added
A better solution would be to use a service specific instance orchestration
© Fraunhofer FOKUS
Slicing, Edge Networks and Flexible Backhaul
Monolithic
Mega-
network of
LTE Customized Slice
Customized Slice
Customized Slice
MME HSS
PGW
MMELGW
8
Customization of the specific functionality
Customization of the specific features
Adding life-cycle agility
Adapting at deployment time to the available infrastructure
Adapting during runtime to new conditions
© Fraunhofer FOKUS
Customizing a core network
Dimensioning
Customization & Flexibility
Security
Quality
Reliability
Performance
Text
Lightweight Control Plane
Access
Contr
ol
Connectivity
Contr
ol
Security
Mobili
ty
QoS
&
Charg
ing
Life-cycle agility
• Initial adaptation to environment
• Runtime adaptations
9
Interconnect and Routing Function
A common bus for all the NF-to-NF
communication
PubSub type of mechanism
NF Repository Function
Based on a “DNS”-like repository
Interconnection Function
A router like function which forwards the requests
Data Layer
Distributed and near real-time storage of
structured data
Data sharing between same type of NF
Data redundancy
Network Functions Granularity
NF NF
NF Repository
Function
NF NF
NF NF
Interconnection and
Routing Function (IRF)
Interconnection
Function
NF NF
State
Information
10
Ensuring that software defined networks reach the same (or better) reliability than physical networks
Efficient state and load sharing mechanisms (core network)
New mechanisms for high availability (core network)
Dynamic spectrum selection (radio management)
Support from the NFV framework (fault management)
Support from the SDN framework (backhaul reselection)
Dynamic device connectivity (device management)
Machine Learning anomalies detection (network management)
Cross topic: Reliability
11
Providing quantitative or comparative evaluations of different network architectures on top of heterogeneous, reproducible network conditions
Support for different syntethic workload or real life workloads replay
Emulation of complex network topologies within a single data center
Monitoring the service KPIs: delays, successful procedures, interrupted sessions
Monitoring of used resources: CPU, memory, storage
Interworking with real devices and radio (if needed)
© Fraunhofer FOKUS
Cross topic: Benchmarking
12
The operator breaks the relationship between
the vendor and the data center provider
Operator becomes an intermediary
Operators departments are split:
Telco – knows the requirements
IT – knows the data centers
The end-to-end services are composed of
multiple pieces coming from one or different
vendors
Ensuring that the software qualities are
supported by the data centers
Proving how much resources of a specific
setup are required
Ensuring that the operator acquires the
appropriate data center
© Fraunhofer FOKUS
NFV is based on a broken business relationship
13
Current performance evaluation mechanisms are based on the individual components
performance evaluations
In NFV, software programs run in parallel, with dynamically allocated resources
Clear separation of performance per-component is not possible
Performance is varying depending on the scaling / resource consumption
Performance of the communication is dependent on the resources availability at both
ends of an interface
© Fraunhofer FOKUS
Redefining the service KPIs
14
System KPIs – quality of the service offered to the subscribers
Call establishment rate, call establishment delay, drop call rate, etc.
Workload – amount of external work which should be handled by a slice
Resources KPIs - how much resources are allocated / are consumed by the service
CPU, memory, storage and network
Life-cycle KPIs – how fast a service can be deployed and upgraded
Cloud-native KPIs – how well a service uses the cloud capabilities
Scaling depending on the load
Transparent reliability
© Fraunhofer FOKUS
New type of service KPIs have to be defined
15
Open5GCore is a pre-standard implementation of the 5G ecosystem
Open5GCore aims to foster 5G development beyond LTE/EPC
More efficient communication for the subscribers (low delay/high capacity)
Providing the users a means to control their environment (automation/reliability)
Providing communication for other markets (Industry 4.0, eHealth, energy, critical)
Open5GCore Rel. 3 is a NON-OPEN SOURCE R&D toolkit enabling:
Deployment of testbed small scale operators
Integration with common radio boxes (LTE, WiFi, 5G prototypes)
Using common phones
Open5GCore is designed for 5G ecosystem R&D needs:
Based on standard components (3GPP, ETSI, IETF, ONF)
Easy to customize, modify and extend
Enabling large input loads & comprehensive monitoring
© Fraunhofer FOKUS
What is Open5GCore?
16
A radical innovative core network for 5G, LTE and WiFi
Open5GCore is an R&D prototype, including features with high industry relevance from the Fraunhofer
FOKUS research activities, based on 3GPP standards (Rel. 11, 12, 13, ...)
The principles of standard alignment, configurability and extensibility have been respected in the overall
architecture and in the specific components implemented
Open5GCore Release 3 features:
Support for LTE and WiFi access networks
Cloud-native core network customized for NFV slices
Seamless elasticity, state sharing and load balancing
Mobile edge computing support
Service oriented data paths
Benchmarking
LTE/5G radio signaling protocol stack
© Fraunhofer FOKUS
Open5GCore Rel. 3
17
Providing quantitative evaluations of different core network implementation architectures
The benchmarking tool and environment include the following functional features:
Flexible and intuitive eNB topology configurations
Flexible subscriber mobility and load patterns –
Support for x1000 emulated subscribers
Support for x10 eNBs running within different processes
Support for S1-MME and S1-U interfaces,
Attachment, detachment and active handover procedures (S1-based)
Monitoring
Quality: Success rate, procedure delay
Performance of the system: compute, storage and network
On demand extensible for different:
RAN topologies or functionality,
mobility and resource patterns
interfaces towards the network
© Fraunhofer FOKUS
Benchmarking on Open5GCore
18
A simple, single contained testbed for enabling the performance measurements
© Fraunhofer FOKUS
Benchmarking: Knowing the truth about the software networks
19
The system has 2 stable states for the different loads due to the current scheduling
system:
the end-to-end procedure delay and drop rate is not directly correlated with the
workload and the resources available
© Fraunhofer FOKUS
Attachments stress test for 4 CPUs / MME
20
The CPU usage is increasing linearly while the load was increasing in stairs
The CPU usage is not linear – specific scalability thresholds should be defined for each
vEPC software
© Fraunhofer FOKUS
CPU and I/O measurements for the 4CPUs load test
21
For less pre-allocated compute resources, the variation in delay is bigger from the
beginning
The system is handling more than half of the 4CPUs load
© Fraunhofer FOKUS
Attachments stress test – 2CPUs for MME
22
Same non linear behavior as in the case of 4CPUs load test
© Fraunhofer FOKUS
CPU and I/O measurements for the 2CPUs load test
23
Almost seamless to the UEs – the state is split per subscriber
A system with 2 control entities with 2 CPUs each behaves less performant than a
system with 1 control entity with 4 CPUs
© Fraunhofer FOKUS
Scaling of the control plane
A second MME is activated
The second MME is stopped
24
The CPU is not correlated with the number of requests
The scaling process introduces a large disturbance in the consumed resources which
should not be confused with a non-scaling disturbance
© Fraunhofer FOKUS
CPU and I/O in the main and in the backup CTRL
28
There is no means to immediately correlate allocated resources and workload with the
system KPIs in an NFV environment
Correlation of system KPIs based on the hardware infrastructure is not trivial
Large scale parallelization is the new overprovisioning and high availability mechanism
Low delay is the hardest to obtain characteristic in NFV (and the main limiting factor)
© Fraunhofer FOKUS
Lessons learned from the initial benchmarking
29
Delay is the only limiting factor
Components can scale, scaling is cheap, sharing of state is cheap-ish
CPU usage is the main limiting factor
CPU level is the one defining the load of a component
Disk access is very limited
Almost no memory was used – 128MB
We take too much care of the memory
Memory is far from being the limited resource
Better usage a larger memory and have less troubles (and troubleshooting)
Uniform low delay on the data plane
Can not be achieved without acceleration libraries
Acceleration libraries need threads
© Fraunhofer FOKUS
Key findings
30
Designed for the current hardware capabilities
Addressing properly acceptable level of performance in the control plane
Machines with a large number of CPUs would be enough
Reducing the procedure delay by proper scheduling of the tasks
Addressing at least a minimal level of performance in the data plane
Bringing data packets to the user space of a virtual machine in a cloud environment
Providing appropriate counterpart for the signaling plane
Addressing large scale parallelization
10-20 data paths programs running in parallel
© Fraunhofer FOKUS
The new software platform
31
In course of the ongoing 5G development, the 5G Playground @ FOKUS continuously
adopts relevant outcomes of experiments and trials …
Based on the existing product base of customers (or third parties);
From experiences with customized versions of the Fraunhofer FOKUS toolkits;
From research collaboration and prototyping new products.
Aiming to maintain the relevance of the testbed environment provided and opening new
market opportunities through raising awareness (demos, whitepapers).
© Fraunhofer FOKUS
Fraunhofer FOKUS 5G related experimentation
Customization, further
development and Integration
New Product Prototype
Existing Customer Product
Ongoing experimentation continuously advances the 5G Playground
32
Multiple live instances of the 5G Playground
that address particular R&D use cases of
Fraunhofer FOKUS
Dedicated core networks (multi-slice);
Low power packet core networks;
Massive device connectivity;
Flexible backhauling;
Multi-data center environments;
Dense wireless environments;
Industrial wireless communication.
Running on top of the cost efficient and off-
the-shelf hardware infrastructures
Providing remote access and experiment
control if needed
© Fraunhofer FOKUS
What is the 5G Playground @ FOKUS
33
High Performance Data Center
Cloud infrastructure providing high computing, storage and networking capacities
Dell Bladecenter (M620, >120 CPU cores, >640GB RAM)
Fully redundant NetApp Metro Cluster (>10TB Storage)
NVIDIA Tesla (C10, C20, K20)
SDN Datacenter copper/fiber switches (1/10/40Gbit/s, HP3800, Pica-8)
Cisco ASA Routers (redundant Internet connectivity)
Mostly Linux OS (Ubuntu LTS) and OpenStack (Juno & Kilo)
Usage
Operational (high availability) shared cloud environment
for multiple live instances of the 5G Playground
Computing and networking platform for experimenters
Toolkit and benchmark hosts
Update servers for remote 5G Playground instances
Public cloud complementing edge computing experiments
© Fraunhofer FOKUS
The 5G Playground @ FOKUS infrastructure
34
Mini and Micro- data centers
Orange Box: Facility edge network computing environment
Supports up to 20 4G small cell base stations
10x Intel NUCs (Ivy Bridge D53427RKE model),
i5-3427U CPU, 16GB of DDR3 RAM, 120GB SSD root disk, Gb Ethernet
D-Link DGS-1100-16 managed gigabit switch with 802.1q VLAN
Lenovo M93P: Desktop core network for each researcher
Intel® CoreTM i5-4570T 2.9GHz 4M (4th generation), 16GB DDR3 RAM,
Gigabit Ethernet, 3x USB 3.0, 1x USB 2.0
Raspberry PI 3: The smallest core network available
1.2 GHz 64-bit quad-core ARM Cortex-A53, 1GB SDRAM,
4 USB 2.0 ports, 100Mbit/s Ethernet
Usage
Pool of heterogeneous edge computing environments for dedicated R&D use cases,
trial experiments and showcases
© Fraunhofer FOKUS
The 5G Playground @ FOKUS infrastructure
35
Permanent Demonstration Environment
Facilitates 5G Playground toolkits and infrastructure in a comprehensive micro-operator
scenario (network slice)
Placed on a vertical setup
(“4G/5G Wall”)
Showcases
Comprehensive environments
Different features within the
same environment
Effects on live testbeds
Edge mobility
Live monitoring information
© Fraunhofer FOKUS
The 5G Playground @ FOKUS infrastructure
36© Fraunhofer FOKUS
What is the 5G Playground made of
• A standard aligned implementation of the ETSI NFV MANO
• Running on top of OpenStack (and soon OpenMANO)
• Providing independent infrastructure slices
• Support for runtime elasticity and fault management
• A large amount of use cases
• Core networks, multimedia, etc.
• Available on github:
• https://github.com/openbaton
37
Scope: Multiple service slices on top of a dynamic federation of heterogeneous compute and storage nodes
© Fraunhofer FOKUS
Experimentation on Network Function Virtualization
NFV in distributed, heterogeneous environments
NFV as a platform for micro-services
38© Fraunhofer FOKUS
What is the 5G Playground made of
• A standard aligned implementation of the ETSI NFV MANO
• Running on top of OpenStack (and soon OpenMANO)
• Providing independent infrastructure slices
• Support for runtime elasticity and fault management
• A large amount of use cases
• Core networks, multimedia, etc.
• Available on github:
• https://github.com/openbaton
• A new approach to device communication and M2M
• Addressing connectivity of a large number of devices
• Connectivity control on top of heterogeneous environments
• Security
• Customized connectivity
• Service capabilities
• Based on standard protocols
• OMA LW M2M, eSIM, etc.
39
The network is becoming more heterogeneous and flexible in terms of flexible areas
Enterprises have to connect their private networks to the carrier networks
For convergent service across distributed geographical areas
For interaction with other service providers (e.g. logistics, manufacturing, safety)
Still a very high security level has to be maintained for the private networks
A large amount of confidential information is available in the private network
A large amount of detailed short living monitoring information
Security Islands Motivation
A continuous flow of connected devices
I4.0 requires that devices change their security zone
while executing the industrial process
Virtualized Network Service / NFV
Provides the ability to replicate functionality within
different network areas, customized to the local
needs
40© Fraunhofer FOKUS
What is the 5G Playground made of
• A standard aligned implementation of the ETSI NFV MANO
• Running on top of OpenStack (and soon OpenMANO)
• Providing independent infrastructure slices
• Support for runtime elasticity and fault management
• A large amount of use cases
• Core networks, multimedia, etc.
• Available on github:
• https://github.com/openbaton
• A new approach to device communication and M2M
• Addressing connectivity of a large number of devices
• Connectivity control on top of heterogeneous environments
• Security
• Customized connectivity
• Service capabilities
• Based on standard protocols
• OMA LW M2M, eSIM, etc.
• Providing an extensive platform for SDN added value features
• Based on standard components (IETF, ONF, etc.)
• Establishment of dynamic data paths
• Backhaul control for dedicated networks
• Deep data plane programmability
• Service Function Chaining
41© Fraunhofer FOKUS
What is the 5G Playground made of
• A standard aligned implementation of the ETSI NFV MANO
• Running on top of OpenStack (and soon OpenMANO)
• Providing independent infrastructure slices
• Support for runtime elasticity and fault management
• A large amount of use cases
• Core networks, multimedia, etc.
• Available on github:
• https://github.com/openbaton
• A new approach to device communication and M2M
• Addressing connectivity of a large number of devices
• Connectivity control on top of heterogeneous environments
• Security
• Customized connectivity
• Service capabilities
• Based on standard protocols
• OMA LW M2M, eSIM, etc.
• Providing an extensive platform for SDN added value features
• Based on standard components (IETF, ONF, etc.)
• Establishment of dynamic data paths
• Backhaul control for dedicated networks
• Deep data plane programmability
• Service Function Chaining
• R&D prototype of mobile core networks beyond 3GPP Release 13
• Support for (5G), LTE and WLAN
• Cloud-native core network for NFV
• Seamless elasticity
• Mobile edge network support
• Service oriented data paths
• Highly customizable (for DCNs)
• Benchmarking and experimentation
42
The 5G Playground was designed from the initial phases for commodity for being
deployed at customer premises
Mirroring the advancements from the FOKUS and the Berlin testbed
Providing a separate isolated testing facility
Including only the interesting functionality from the comprehensive environment
Customizing the test environment for the specific requirements
© Fraunhofer FOKUS
Clone and customize your own 5G Playground
© Fraunhofer FOKUS
Your
Premises
43
Local Setup of the 5G Playground
Infrastructure
Local Radio Access Network
Apps Apps
Apps
Customized Network Configuration
Live Experiment Results
Remote Maintenance and Upgrades
End-to-End use cases federation
44
Testbed at customer premises
Cloning and customizing the testbed at
customer premises provides the needed
know-how boost in the applied 5G
research
Own hardware
Own administration
No travel costs
Demo available when needed
© Fraunhofer FOKUS
A remote testbed access is more expensive than a local one
Remote testbed access
Provides initial demonstrations of a
specific solution, with limited know-how
transfer
Using FOKUS hardware
Employing a FOKUS administrator
Traveling to FOKUS
Sharing the environment
45
5G Berlin currently consists of 4 central nodes (Communication & Datacenters) and 2 experimentation sites (Fraunhofer FOKUS & HHI).More sites joining until end of 2016.
5G Berlin testbed sites
www.5gberlin.de
TUB
FOKUS
HHI
46
Development of a live outdoor environment for 5G
Based on the newest advancements towards 5G radio
Distributed edge computing nodes handling core network functionality
Customized front-haul and backhaul towards dedicated edge networks
Real communication conditions
Public available and new apps and services
© Fraunhofer FOKUS
5G Berlin urban trial platform
47
Providing 5G access, 5G core, Xhaul with SDN/NFV/MEC service platforms within a single place
Following the NGMN requirements for 5G
Based on the combined experience of Fraunhofer FOKUS and Fraunhofer HHI
Uniting the complete stakeholders value chain for 5G
A unique combination of the current 5G advancements
A place for operators, vendors and integrators
to develop and test new 5G concepts
© Fraunhofer FOKUS
5G Berlin
48
5G World federated testbed infrastructure
Location Independent Infrastructure Management
Internet Backhaul
Satellite Backhaul
5G World currently consists of
49
For further information, technical questions, licensing and
pricing requests, contact us at [email protected]