Virtual Network Services As Enabler of
Dynamic Application-Aware Traffic Engineering
Masato Tsuru
Network Design Research Center, Kyushu Institute of Technology, Japan
Ultimate Goal (or Problems already faced...)
Support huge network traffic sustainably
From YouTube, software updates, to Cloud, M2M,
ITS, Smartgrid, ... in cost and energy efficient ways
Enable/support diverse, innovative network
applications (in diverse environments)
Distributed, Collaborative, Cyber-physical, Wireless,
Distributed, Collaborative, Cyber-physical, Wireless,
Mobile, Ad hoc, ...
Diverse demands and conditions: e.g., Throughput
v.s. latency, Security, Mission criticalness, ...
Resilient for extra. situations
Disaster or cyber-attack damage tolerance and fast
recovery (We cannot assume perfect protection)
Internet Traffic Monitoring in Japan: Still Growing
3
Mobile Traffic Growth Forecasts ... exponential?
(from Report ITU-R M.2243: Assessment of the global mobile broadband deployments and forecasts for International Mobile Telecommunications))))
Bandw idth narrow S a te l S a te l S a te l S a te lli tei tei tei te Bandw idth narrow S a te l S a te l S a te l S a te lli tei tei tei te
IF multiple different networks can be used for a single
transfer task in an integrated fashion,
the efficiency can
be increased
and/or
the cost decreased
Network Access Infra. Diversity: Still Growing
5 Population density broad Optica l /ADS L VO IP WIFI WIFI WIFI WIFI WIFI WIFI Cellular phone C HAT E -Mai l W EB AUDI O S TREAM ING IP TV UP DATE DOW NL OAD urban rural rural S tore-carry-forward scheme based network infrastructure Population density broad Optica l /ADS L VO IP WIFI WIFI WIFI WIFI WIFI WIFI Cellular phone C HAT E -Mai l W EB AUDI O S TREAM ING IP TV UP DATE DOW NL OAD urban rural rural S tore-carry-forward scheme based network infrastructure
Where and How Can We Solve the Problems ?
Applications/Users (L7)
–
Save an unnecessary use of resource
–
Life can change; But do not discourage economy
Networking (L2-6) -- It's OUR field.
–
Allocate
physical resources
efficiently for sufficient
–
Allocate
physical resources
efficiently for sufficient
functionality and performance of
each application
–
TE (Traffic Engineering) in a widest sense
Physical Communication Media (L1)
–
Increase network capacities by deployments or new
complex technologies
Networking Solutions
Effective and Efficient Resource Sharing (TE)
– More application-aware and physical media-aware dynamic, flexible, adaptive allocations are needed
– 3 New technology trends (involving each other)
1.
Asynchronism (non-realtime/non-interactive)
– Typically seen in DTN and CCN (ICN)
– Freedom of Time and Space for Optimization: Store and scheduling, Multi-network-path, Prefetch and cache,
7
scheduling, Multi-network-path, Prefetch and cache,
Information Coding, Further In-network processing, ...
2.
Multiple time and space-scale control.
E.g.,
– Packet-level Scheduling and processing
– Flow Scheduling for simultaneous competition
– Synthetic mechanism for traffic peak shift in large
– Phase change for extraordinary situations (less resources)
3.
User/Social Activity Interaction
Two Players in a Basic Model
InP
InP
InP
InP
(Infrastructure Provider)
–
Provide physical resources: Regional and Global,
Backbone and Wireless, ..
–
Like a public service; Suffer flat rate
–
Know and control physical resources
SP
SP
SP
SP
(Service Provider)
SP
SP
SP
SP
(Service Provider)
–
Contents and/or applications: Google, Yahoo!, Amazon,
Apple, Skype, navigation services, game services, ...
–
Cause huge and diverse traffic
–
Know and control applications and users
Architecture/Model for Design/Implementation
To allow diverse design choices
– Trade-off on Performance, Efficiency, Resiliency, Fairness, .
– Strict/Probabilistic optimization, Heuristics, Game-theoretic, ...; Centralized/Decentralized
– New application interfaces and user interaction
To solve real world requirements
9
– Business Model, Standardization, Regulation,
– ID/Locator separation, Distributed Security/AAA
Can Network Virtualization Help us?
– Virtual Network (VN) per app/service differentiated
– VN for extraordinary situation change
– Incremental Deployment, Extensibility, Programmability, ..
Diverse Services/Users
アプリ アプリ アプリ アプリ
TCP/IP = A single Virtualized Network by integrating M resources
Traditional Internet
アプリ アプリ アプリ アプリ
Provide N diverse Virtual Networks
Flexible Internet
Do Optimal Matching
Save Demand Diverse Services/Users
1 N 1 N
Network Virtualization Enabling Application-Aware TE
Shif to
Diverse Network Infra/Resources
通信環境 通信環境 通信環境 通信環境
by integrating M resources
通信環境 通信環境 通信環境 通信環境
by combining M resources
Be Flexible, Extensible, Robust, and Open
Increase Capacity Matching
Assume (almost) realtime, bidirectional data exchange along a single (almost) stable path between fixed end-hosts
1 M
Diverse Network Infra/Resources
M 1
VN for SP (1)
Service Provider ((((SP))))
•Has a proprietary VN to provide its services to end users
•The VN is provided by VNP
Virtual Network Provider ((((a Reliable Middle
Layer)
A Middle Layer Model (Virtual Network
Provider)
VN for SP (2) VN for SP (3) 11 InP (1) InP (2) Layer)•Coordinator between SPs and InPs
•Provide a SP's VN in cooperation with InPs (Abstraction, Integration. Separation,..)
•Operated by a union of InPs??
Infrastracture Provider (InP)
•Provide resources (network, storage, computation) to SPs via VNP
Our Research Plan
Application-Aware New Generation TE
– Huge traffic, Diverse apps, Resiliency, and Fairness
– By introducing Asynchronism, Multiple scale control, and User Interaction
– Intra-VN TE and Inter-VN TE; the latter is more challenging
Virtual Network Service Architecture enabling new TE
– A Middle Layer Model -- A proof-of-concept development of– A Middle Layer Model -- A proof-of-concept development of Control and (perfSONAR-based) Management planes
– InP: OpenFlow + In-Network servers (storage and proc.)
• OpenFlow has a potential of very flexible TE
• We use Trema-based controler (may be easy to use?:-)
– SP: Use case of realtime and non-realtime applications
Experimental Analysis & Evaluation
– JGN-X and OpenFlow testbed on itElapsed Time B a n d w id th A ll o c a ti o n 100 Normal parallel transfer T ra n s mi s s io n C o mp le ti o n T ime [s ] average
A Simple Flow Scheduling (Just a Serialization)
13 1st Flow's and Averaged Transmission Completion Times are reduced
Wait Elapsed Time
B a n d w id th A ll o c a ti o n 100 Serialized transfer T ra n s mi s s io n C o mp le ti o n T ime [s ] average
Preliminary estimation of the effect of a simple scheduling:
When sender A wants to send a new file to receiver B, A
When sender A wants to send a new file to receiver B, A
When sender A wants to send a new file to receiver B, A
When sender A wants to send a new file to receiver B, A
declares the file size and
declares the file size and
declares the file size and
declares the file size and WAITS IF
WAITS IF
WAITS IF
WAITS IF
–Some other shorter flow is running/wants to run on the bottle neck Some other shorter flow is running/wants to run on the bottle neck Some other shorter flow is running/wants to run on the bottle neck Some other shorter flow is running/wants to run on the bottle neck link for A's new flow.
link for A's new flow. link for A's new flow. link for A's new flow.
A Simple Flow Scheduling (Simulation settings)
Simulation Model
–At time 0, each sender (at Leftsender (at Leftsender (at Leftsender (at Left----side)side)side)side) wants to send TWO files to different randomly chosen receivers (at Rightreceivers (at Rightreceivers (at Rightreceivers (at Right---side)-side)side)side). So EIGHT flows compete in total.
–Each file has a random size(50[Mbit]~500[Mbit])
–Each flow traverse a single path which is randomly chosen among the shortest paths from the sender to the receiver..
50 60 70 に よ る 実 効 転 送 レ ー ト に よ る 実 効 転 送 レ ー ト に よ る 実 効 転 送 レ ー ト に よ る 実 効 転 送 レ ー ト [M b p s]
A Simple Flow Scheduling (Results of 100 trials)
Effective Rate = Transmission completion time / File sizeX: Effective Rate by the normal parallel transfer
Y: Effective Rate by a simple (serialized) scheduling
15 0 10 20 30 40 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 グ ル ー プ グ ル ー プ グ ル ー プ グ ル ー プ S C Hに よ る 実 効 転 送 レ ー ト に よ る 実 効 転 送 レ ー ト に よ る 実 効 転 送 レ ー ト に よ る 実 効 転 送 レ ー ト 通 常 転送による実効転送レート 通 常 転送による実効転送レート通 常 転送による実効転送レート 通 常 転送による実効転送レート[Mbps]
VY C2 C3 トポロジ情報DB 資源情報DB ミドルレイヤ SS C1 C2 C3 S1 トポロジ情報DB 資源情報DB サービスレイヤ
Middle Layer Model: A proof-of-concept development
•Resource View Mapping (DBs) •SP-VNP, VNP-InP Interface (API) --Topology, Link capacity •Endhost mapping •Use case app--X1 X2 X3 Y1 Y3 Y2 Z1 Application server Application client S1 C1 C2 C3 InP X InP Y InP Z トポロジ情報DB 資源情報DB トポロジ情報DB 資源情報DB トポロジ情報DB 資源対応表DB VX VZ C1 S1 インフラレイヤ
•Use case app --multicast delivery
Middle Layer Model: A proof-of-concept development (2)