lifting the boundary between sdn and nfvnv/nfvsdn-symposium/nfvsdn-sympo2014-04-nakao.pdf · global...
TRANSCRIPT
Lifting the boundary between SDN and NFV
SDN&NFV Symposium Aki Nakao
The University of Tokyo [email protected] / http://nakao-lab.org
1
Global Trends:
Synergy between Networking and Computing
2
SDN NFV
We posit that the boundary may become blurer
3
FLARE Architecture Vision: Software-Defined Data-Plane Enables Fusion of SDN &NFV
Near Future
Data Plane API
Commodity Hardware (x86, ManyCore NetProc)
Monolithic Orchestrator
Orchestrator
No AssumpBon of IP Network
NBI
Data Plane Processing (SDN Switch + NFV)
Control PlaKorm
SAL
SoMware Defined Data Plane
Present
Northbound Interface (NBI)
Service ApplicaBons
Control PlaKorm
Southbound Interface (SBI)
SDN HW OpenFlow JUNOS OnePK
Service AbstracBon Layer (SAL)
NFV Control PlaKorm
Orchestrator
NFV
Hardware SoMware
Commodity Hardware
SDN SoMware Switch
SoMware
Commodity Hardware
Data Plane Elements(DPE)
AssumpBon of IP Network
DPE
ITU-T Y.3300 (Y.SDN-FR) “Framework of software-defined networking,”
4
5
Comparison among Data-Plane Technologies (Resource Isolation, Ease of Programming, Power Consumption and Performance)
Many-Core Network Processor (Linux)
Many-Core Network Processor (Assembly)
DPDK x86
ASIC/ Silicon
NetFPGA
Ease of Programming
+ ー + ー ー
Performance + + + + +
Multiple Logic Isolation
+ + ー ー ー
Low Power Consumption
+
+
ー
+
+
6
FLARE Node Implementation
36-72 cores (upto 100-200 cores in future)
(board designed by NakaoLab)
x86 Processor
Many Core Processor
The University of Tokyo Confidential
Hierarchical Resource Management • General Purpose Processor(s) • Network Processor(s) • ...and more types of processors
7
FLARE-EX Prototype (32x10Gbps)
8
9
Control-Plane Elements
Network Applications
Programmable Data-Plane Elements
Applications (NFV)
Control Plane
Data Plane
North-Bound Interface (NBI)
E.g. OpenFlow Switches
Packet Process
Packet Process
Publish API
Packet-In Processing Offloading in Software Defined Data Plane
10
Toy-Block Networking
ToyBlock Networking GUI
Resource Usage Optimization in Software Defined Data Plane Elements
12
• H/W OpenFlow is One-size-fits-all
• Software-Defined Data-Plane enables tailored optimization for only functionalities operators require • Software-Defined OpenFlow 1.3 on FLARE
• Optimized for a small number of “kinds” of flow entries • Supports 1M flow entries for multi-10G switching • Low-Cost / Low Power
So#ware Defined OFS1.3 on FLARE Performance vs. CPUs
1.9
0
2
4
6
8
10
12
1 2 3 4 5 6
Gbp
s
#CPUS
pkt_size=1514B, 1 million flow entries
13 Throughput linearly scales up as the number of CPU cores
So#ware Defined OFS1.3 on FLARE Performance vs. Flow Entries
2.4 1.9 1.88
0 0.5 1 1.5 2 2.5 3
Gbp
s
#flow entries
pkt_size=1514B, one CPU
14 Throughput stays flat regardless of the number of flow entries
Customized Actions • Some of NFV functions could be offloaded in data plane elements
E.g., OpenFlow Extended Actions
15
OpenFlow Element
Custom Actions
16
17
Window-based Arbitrary Bit Matching
Arbitrary bit matching as in openflow pattern matcher is costly due to expensive memory operation per packet
window
Set a window to minimize per-packet memory operations Improve performance while keeping flexibility
Leon Lee, Ping Du and Akihiro Nakao, "Ouroboros: SDN Beyond Flow-Tuple Matching," IEICE NS Technical Report, Mar. 2013
18
Window-Based POF Bit Matching
Leon Lee, Ping Du and Akihiro Nakao “Ouroboros: SDN Beyond Flow Tuple Matching” IEICE Technical Report, Mar. 2013
The University of Tokyo Confidential
Application-Oriented Southbound Interface (SBI)
19
From Network Operator Point of View Network Operation based on “Flows” Application Developers and End Users Point of View Network Operation based not on Flows but on Apps
ApplicaBon Specific TE (patented)
20
SDN/NFV Incubation Program 3 days of intensive discussion with NTT Labs,
NTTCom, KDDI, Freebit, etc. on resolving real issues
Software Eats Everything…
Software Defined Data-plane may lift boundary between SDN and NFV
22