Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Converging SDN and NFV at the Network Edges: The lightMANO ApproachRoberto RiggioCHIEF SCIENTISTFBK CREATE-NET, Italy
COST ACROSS Workshop "Management & Control for Reliable Softwarized Networks"Genova, Monday 4th, September 2017
Mission:▪ Softwarization of Digital Infrastructures▪ Cloudification of ICT Resources and Services ▪ Robust, Secure and Application-centric Infrastructures
Research Lines:▪ Open platforms for IoT▪ Distributed & autonomic cloud computing▪ Future mobile & core networks
DIVINE: a facility for Next Generation Internet experimentation
CREATE-NET Center: Digital Infrastructures
OpenIoT: Open platforms for IoT
Toward a Tactile Internet ▪ Autonomous systems with Cognitive IoT ▪ Decentralised IoT infrastructures with blockchain
Proximity Cloud Computing:▪ Resource optimization in highly distributed Cloud
infrastructures (fog computing)▪ Economics and business models for decentralized
and multi-stakeholders environments
DISCO: Distributed & Autonomic Cloud Computing
One (Programmable, Secure, Sliceable) Network:▪ End-to-end slicing in 5G networks▪ Lightweight Management and Orchestration▪ SDN & NFV for network control/management and service
orchestration
FuN: Future Networks
▪ A Distributed, Virtualised and Programmable Digital Infrastructure
▪ 5G, Fog and IoT experimental activities▪ Playground for FBK researchers for NGI scenarios
DiVINE: a facility for Next Generation Internet experimentation
The Pendulum is Swinging Back!
Focus is moving back to the edges of the network
Mobile Edge Computing, Fog Computing, Dew Computing, …
What is Edge (or Fog, Dew) Computing?
Cloud
Fog/MEC/Dew
Devices
❏ Cloud-based computing, storage, and networking at the edges of the networks❏ Computational locality❏ Low Latency❏ Reliability❏ Security
“the cloud computing is dead, the intelligence/processing is going down close to the things”
Peter Levine (Andreesen&Horowitz) at Gartner Keynote
Factory - Energy Media & entertainment e-Health Automotive
Business
applica
tion lay
erVertical
M(V)NOsEnterprise
OTT
Tenant 1Tenant 2
Virtualized RANFront-/
BackhaulEdge Cloud Virtualized Transport
Network Virtualized Mobile Core Network
Virtualized Cloud
Wholesale Operator
Technica
l Enabl
ers
Infrast
ructure
Layer
5G: One (Programmable, Sliceable, Secure) Network
API
VIRTUALIZATION
Network Function Virtualization❏ Network Function Virtualization is one of the pillars of network
softwarization❏ Designed with a cloud-computing / datacenter mentality
❏ Virtualization (computing, network, storage)❏ Network functions implemented in software❏ General purpose hardware/software❏ Elastic provisioning
❏ Several Management and Orchestration frameworks❏ ETSI MANOs (OpenMANO, OPNFV, OSM, ...)❏ Commercial (Amazon, Azure, Ubuntu, ...)
ETSI MANO ArchitectureNFV Infrastructure Layer:❏ The environment where VNFs are deployed
Management and Orchestration Layer:❏ Virtual Infrastructure Managers❏ VNF Managers❏ Orchestrator
Network Service Layer;❏ Virtual Network Functions❏ Element Manager❏ OSS/BSS
NFVI
Hardware ResourcesComputing Hardware
Storage Hardware
Network Hardware
Virtualization Layer
Virtual ResourcesVirtual
ComputingVirtual
StorageVirtual
Storage
EM 1 EM 2 EM 3
VNF 1 VNF 2 VNF 3
OSS/BSS
Virtual Infrastructure
Manager(s)
VNF Manager(s)
NFV Orchestrator
❏ Virtual network functions can be deployed at the most appropriate location: in the cloud, in central offices, or at the customer premises
❏ Not a new concept:"Network Functions Virtualization – Introductory White Paper, 2012"❏ Rationale:
❏ Security: end-to-end security requires some functions to be implemented at the customer premises❏ Performance: some applications may require local computation in order to reduce latency❏ Resiliency: scattered facilities may rely on the public Internet for their interconnection. Dedicated WANs may not
be economically viable or feasible in all situations
Distributed NFV
Customer Premises Central Office Cloud
VNFs VNFs
VNFs
Distributed NFV
Challenges❏ Management and Orchestration ❏ Interplay between NFV and SDN❏ Heterogeneity❏ Software engineering and DevOps
Management and Orchestration
Site NCustomer Premises
Cloud
Physical Resources
Virtualization
Virtual Resources
MANO
VIMs
VNF-M
Orchestrator
Site 1
Physical Resources
Virtualization
Virtual Resources
❏ Tens of thousands or even millions of sites❏ Residential❏ Industry 4.0❏ IoT / Smart cities
❏ Edge resources are constrained, need to make the best use of them
❏ Centralized MANO frameworks limitations:❏ Heavyweight virtualization solutions (VM)❏ Complex infrastructure controllers (OpenStack)
❏ MANO Components not in the same LAN❏ Increased latency❏ Intermittent or lack of connectivity❏ Upgrading the WAN is not always economically
viable or feasible
Management and Orchestration
Site 2
Physical Resources
Virtualization
Virtual Resources
lightM
ANO
Site N
Physical Resources
Virtualization
Virtual Resources
lightM
ANO
Site 1
Physical Resources
Virtualization
Virtual Resources
lightM
ANO
❏ With tens of thousand or millions of NFVI PoPs there could be 10x/100x VNFs
❏ Edge services may be volatile, requiring fast provisioning and decommissioning
❏ Traditional ETSI MANO implementations can not cope with this scenario
❏ VNF placement/scaling to account for scattered and distributed deployments
❏ Move (part of) the orchestration logic to the network edges
SDN & NFV Beyond Simplistic Models❏ SDN & NFV have largely evolved independently from each and only recently we are starting to
witness the first convergence attempts❏ Cloud computing platforms such as OpenStack have not been designed with NFV Management
and Orchestration in mind❏ Networking API limited to simplistic VLAN–based service function chaining models❏ Virtualization and abstractions primitives available in state–of–the–art SDN platforms not yet
exposed to the service orchestrator
SDN & NFV Beyond Simplistic Models❏ An LTE Network deployed as a network service❏ Reuse the same substrate for different network services❏ Multiplexing gain: computing, power, sites❏ Dynamic and flexible provisioning of resources
Backhaul
Fronthaul
Mobile Edge Cloud
Mobile Edge Cloud
S-GW
P-GW
NFVI-PoPFronthaulRRH
RRH
eNBeNB
SDN & NFV Beyond Simplistic Models❏ Flexible middlebox provisioning ❏ Content caching, augmented reality, localized processing
Backhaul
Fronthaul
Mobile Edge Cloud
Mobile Edge Cloud
S-GW
P-GW
NFVI-PoPFronthaulRRH
RRH
eNBeNB
CDN
Handle Heterogeneity❏ How is the distributed NFVI supposed to look like?
❏ Need to mix physical and virtual network functions❏ Need to interface with legacy systems and with heterogeneous infrastructures
❏ Highly heterogeneous environments❏ Different access technologies: Wi-Fi, LTE, LoRA, ...❏ Legacy technologies: netconf, SNMP, CLI, ...❏ Mix physical and virtual network functions
❏ "Compile" network service descriptors for different "targets", e.g. netconf, OpenFlow, JuJu, etc.
Handle Heterogeneity❏ Multi-domain NFV in satellite networks❏ Combines Physical and Virtual Network
Functions❏ Legacy network protocols❏ Non ETSI MANO compliant VIMs❏ Interdomain service function chaining❏ Mix of SDN controllers and WAN
controllers
Software engineering and DevOps❏ Telcos are already moving from configuring network service to programming network services❏ ETSI MANO frameworks already rely on declarative languages for
❏ Service on-boarding❏ Service scaling❏ Fault management and monitoring
❏ DevOps model, creating a tight creation, testing, deployment and operation lifecycle❏ Challenges:
❏ Virtualized nodes mirrors their hardware equivalent❏ Opportunities
❏ Re-factor of functionalities❏ Micro-service model
❏ Decomposed network functions into their individual components and scale them independently
❏ Network functions can be decomposed and re-factored❏ 3GPP is standardizing several functional splits❏ Splits could be adapted to network conditions (e.g., day/night)
Software engineering and DevOps
PHY
DU
PHY
DUBackhaul
Fronthaul
Mobile Edge Cloud
Mobile Edge Cloud
RRC
PDCP
RLC
MAC
CU
S-GW
P-GW
NFVI-PoPFronthaul
lightMANO 0.1❏ Converged SDN & NFV Multi-access
Network Operating System❏ Lightweight virtualization❏ Unified control and coordination for
heterogeneous RATs (Wi-Fi and LTE)❏ Programming abstractions
❏ State management ❏ Global network view❏ Network reconfiguration❏ Service Function chaining
❏ COHERENT SDK❏ Multi-tenant
5G-EmPOWER OS
Hardware Abstraction Layer
COHERENT (Python/REST)
EmPOWER Runtime
LightMANO
capwap openrannetconf openflowBackhaul Controller
Intent Compiler
lvap/lvnf-p
Path Computation Element
Northbound Interface (REST)
SLA Monitor Placement Engine VNF Catalog
TOSCA Parser
NFVI
VIM +
VNF-M
Orch
estra
tor
GUI
AAA
Infrastructure
EPCEdge Node
Lightweight virtualization solutions
VNF
Hardware
Host OS
Hypervisor
Guest OS
Libraries
VNF
Guest OS
Libraries
VNF
Hardware
Host OS
Libraries
VNF
Libraries
Hardware
Host OS
Docker
Virtual Machines Containers Kernel bypass
VNF
Libraries
VNF
Libraries
Lightweight virtualization solutions: Comparison
Pro Cons
VM ❏ Complete isolation between VMs❏ Common practices more consolidated
❏ Dedicated OS instance for each VNF❏ Lower density
Docker ❏ Single kernel shared between containers❏ Higher density
❏ Lower isolation❏ Lower portability
Kernel Bypass ❏ Fast to boot, low memory footprint❏ High data-path performances
❏ Very low isolation❏ High entrance barrier
❏ A framework for implementing packet processing pipelines❏ A Click router is built starting from small components called "Elements"❏ Each Element implements a simple operation, e.g. decrease an IP packet's TTL field❏ Connections between elements represents possible paths that can be followed by packets
❏ Lowers the entrance barrier to Kernel Bypass technologies (e.g. DPDK)❏ LightMANO uses Click to implement fast-data path operations
❏ User-level, Kernel, DPDK-accelerated
FromDevice (eth0) Counter Discard
The Click Modular Router
❏ Encapsulate a wireless client state❏ Enables seamless mobility as VNF migration❏ Implements the 802.11 state-machine❏ Sort of similar to a Virtual Network Function for processing Wi-Fi frames
LVAP Host LVAP Host LVAP Host
LVAP LVAP LVAPLVAP
The Light Virtual Access Point
❏ A generalization of the LVAP concept❏ An abstraction for NFV management and orchestration❏ Network services as a composition of modular and reusable blocks❏ Custom packet processing on a precise portion of the flowspace❏ Fast deployment and migration
Generalizing the LVAP: The Light Virtual Network Function
DPI
LVAP GW
tp_dst=80
*
LVAP Host
LVNF Host
LVNF HostOF Switch
The COHERENT SDK from 10000 ftLVAP Migration (Handover)
# Handover client to AP with best RSSIlvap.blocks=self.blocks(). sortByRssi(lvap.addr). first()
LVNF Deployment and chaining
# Click scriptVNF="in_0 -> d::WifiDupeFilter() -> out_0"
# Create Imageimg = Image(vnf=VNF)
# Spawn LVNFdupe = self.spawn_lvnf(image=img, cpp=cpp)
# Chainig LVAP to Duplicate Filteringlvap.ports[0].next[""] = dupe.ports[0]
# Chaining Duplicate filter to DPIdupe.ports[0].next["tp_dst=80"] = \ dpi.ports[0]
Deploying LightMANO
Docker
RAN Node (LTE)
Docker
RAN Node (Wi-Fi)
Edge Node
Docker
5G-EmPOWER Ryu LightMANO
Edge Node
DockerEdge Node
Docker
Squid
LightMANO InterfacesDatapath
Legacy VNF
LVAP Host
Docker Container
LVNF Host
EPC
Implementation Details❏ RAN Nodes (Wi-Fi)
❏ PC Engines APU2 (AMD Jaguar, 4GB RAM, 64GB Storage)
❏ Ubuntu 17.04 Server❏ Docker + LVAP Host
❏ RAN Nodes (LTE)❏ Intel NUC (Intel Kaby Lake, 32GB RAM, 512GB
Storage)❏ Ubuntu 17.04 Server❏ Docker + srsLTE
❏ Edge Nodes❏ Soekris 6501❏ Intel NUC❏ Raspberry
❏ Kubernetes
❏ Uplink UE traffic received by an eNB over its air interface is encapsulated into a GTP packet and then delivered to the Serving Gateway (S-GW) over a UDP/IP socket
❏ This GTP tunnel is terminated at the S-GW where a new GTP tunnel to the P-GW is created❏ The P-GW removes the GTP header and forwards the UE traffic to its intended destination❏ The same process happens in the reverse direction
The LTE Network Primer
eNodeB
EPC
S-GW P-GW
S1 S5
UE IP UE IPGTPUDPIP
Mobile Edge Caching
GTPEncap/Decap
Encapsulated IP Traffic (UE)
S1 Traffic❏ The Squid Proxy needs access to the UE IP traffic in order to implement caching
❏ An LVNF implementing stateful GTP encap/decap is deployed in an LVNF-Host by LightMANO
❏ Traffic to/from the UE is steered through the LVNF❏ Can be applied to other edge service (augmented reality, IoT, etc.)
EPC
Docker
Squid
Conclusions
❏ Multi-access Edge Computing scenarios require a purpose-build Management and Orchestration solution❏ Scattered Deployment❏ Tens of thousands of NFVI PoPs❏ 10x or even 100x VNFs❏ Constrained execution environments
❏ LightMANO Features❏ Converged SDN & NFV platform ❏ Unified API for wireless access control and coordination❏ Lightweight computing and networking virtualization❏ Proof-of-Concept Implementation
What I did not talk about (partial list)❏ Security
❏ New threats models❏ Multi-tenancy
❏ Orchestration must be slicing aware❏ Policy
❏ Role of regulators❏ New business models
❏ Who pays for the infrastructure?❏ ...