Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation

Transcription

1 Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation George N. Rouskas Department of Computer Science North Carolina State University Joint work with: Ilia Baldine (RENCI), Rudra Dutta, Anjing Wang, Mohan Iyer (NCSU) Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.1

2 In Search of Next Generation Internet Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.2

3 In Search of Next Generation Internet Early Pioneer Work NSF GENI NewArch (DARPA) SIGCOMM FDNA NSF FIND Pouzin Society Euro!NGI Euro!4WARD Euro!FIRE Asia Future Internet Forum Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.2

4 Outline Motivation: The Layering Conundrum Context: The Clean-Slate Debate Net SILOs: The Story So Far Summary and Demo Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.3

5 The Internet is Broken! Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.4

6 The Internet is Broken! Fixed layer architecture is outdated App App App Transport Network Data Link Physical Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.4

7 Challenges with Current Protocol Stack 1. Evolution: function-heavy protocols with built-in assumptions 2. Experimental implementations: require kernel modifications 3. Cross-layer design: lack of inter-layer interactions/controls Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.5

8 Protocol Evolution: Transport Several distinct functions: identify application endpoints (ports) e2e congestion control multi-homing (SCTP) reliability semantics (TCP, RDP, SCTP, etc) evolution of individual functions affects entire transport layer Built-in assumptions about IP addresses transition to IPv6, support for mobility difficult Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.6

9 Experimental Implementations Experimental network designs crucial for: gaining insight understanding protocol operation discovering new knowledge firmly rooted in physical world Implementations on commodity HW/SW remain challenging: require modification of OS kernel involve significant expertise limit ability to play with network stack Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.7

10 Cross-Layer Design Cross-layer design a major research theme over last decade: wireless networks TCP congestion control optical networks (next slide) Adoption of ideas in operational networks quite slow: no interfaces for inter-layer interactions/cross-layer controls lack of experimental work reliance on simulation with invalid assumptions Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.8

11 Software Defined Optics Optical substrate can no longer be viewed as black box Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.9

12 Software Defined Optics Optical substrate can no longer be viewed as black box Collection of intelligent and programmable resources: Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.9

13 Software Defined Optics Optical substrate can no longer be viewed as black box Collection of intelligent and programmable resources: optical monitoring, sensing mechanisms amplifiers, impairment compensation devices tunable optical splitters configurable add-drop programmable mux-demux (e.g., adjust band size) adjustable slot size Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.9

14 Cross-Layer Interactions Impairment-aware RWA and network design Placement of optical sub-systems (converters, amplifiers, regenerators) Traffic grooming Inter-layer QoS and traffic engineering Optical layer multicast Multi-layer failure localization and recovery App App App Transport Network Data Link Physical Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.10

15 The Internet is Doing Just Fine, Thank You! Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.11

16 The Internet is Doing Just Fine, Thank You! Biological metaphor: mutation and natural selection [Dovrolis 2008] Evolutionary designs: more robust, less expensive Mid-layer protocols must be conserved not ossified innovation at lower/upper layers of architecture Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.11

17 Clean-State Initiatives Design the Internet from scratch: Research in new network architectures FIND (US) security, addressing, protocols, routing, economics, theory, large-scale experimental facilities GENI (US) control plane, virtualization, slicing, Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.12

18 Role-Based Architecture (RBA) [BFH 2003] New abstraction: organize protocols in heaps, not stacks Richer interactions among protocols flexibility Require new system-level implementations A B C RSH 1 RSH 2 RSH 3 Payload Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.13

19 Recursive Network Architecture (RNA) [TP 2008] Meta-protocol: generic protocol layer with basic services Each layer in stack appropriately configured instantiation Allows reuse, cleaner cross-layer interactions, dynamic composition MP!4 MP!3 MP!2 MP!1 Physical Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.14

20 Recursive Network Architecture (RNA) [TP 2008] Meta-protocol: generic protocol layer with basic services Each layer in stack appropriately configured instantiation Allows reuse, cleaner cross-layer interactions, dynamic composition MP!4 MP!3 MP!2 MP!1 Wireless Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.14

21 Recursive Network Architecture (RNA) [TP 2008] Meta-protocol: generic protocol layer with basic services Each layer in stack appropriately configured instantiation Allows reuse, cleaner cross-layer interactions, dynamic composition MP!4 MP!3 MP!2 MP!1 Optical Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.14

22 Layering As Optimization Decomposition Protocol layers integrated into mathematical framework [CLCD 2007] [LSS 2006] Global optimization problem: network utility maximization Decomposition into subproblems layering optimal modules (protocols) map to different layers interfaces between layers coordinate the subproblems Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.15

23 Layering As Optimization Decomposition Clean-state optimization layered network architecture optimal layering TCP/IP stack various representations of optimization problem different layered architectures (loose) coupling among layers cross-layer considerations Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.15

24 Our View Internet architecture successful in accommodating change Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.16

25 Our View Internet architecture successful in accommodating change But: current practice of patches/tweaks cannot continue forever Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.16

26 Our View Internet architecture successful in accommodating change But: current practice of patches/tweaks cannot continue forever New architecture must be designed for adaptability/evolvability Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.16

27 Our View Internet architecture successful in accommodating change But: current practice of patches/tweaks cannot continue forever New architecture must be designed for adaptability/evolvability New architecture must preserve/generalize layering Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.16

28 Our View Internet architecture successful in accommodating change But: current practice of patches/tweaks cannot continue forever New architecture must be designed for adaptability/evolvability New architecture must preserve/generalize layering SILO objective: Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.16

29 Our View Internet architecture successful in accommodating change But: current practice of patches/tweaks cannot continue forever New architecture must be designed for adaptability/evolvability New architecture must preserve/generalize layering SILO objective: The goal is not to design the next system, or the best next system, but rather a system that can sustain continuing change Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.16

30 SILO Architecture Highlights Building Blocks: services of fine-grain functionality Design Principles: 1. Generalize traditional layer stack 2. Enable inter-layer interactions: knobs: explicit control interfaces 3. Design for change: facilitate introduction of new services 4. Separate control from data functions Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.17

31 Generalization of Layering Silo: vertical composition of services preserves layering principle Per-flow instantiation of silos introduces flexibility and customization Decoupling of layers and services services introduced at point in stack where necessary Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.18

32 Silos: Generalized Protocol Stacks App App App S1 S1 S2 Cross! Service Tuning S4 S3 S3 Silo & Service Mgmt S5 S6 S6 Knobs S7 S8 S8 S7 S9 Composability Constraints Physical Layers Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.19

33 Inter-Layer Interactions (1) Knobs: explicit control interfaces adjustable parameters specific to functionality of service enable info exchange among services Algorithms may optimize jointly the behavior of services in a silo data interface Service knobs Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.20

34 Inter-Layer Interactions (2) Upward information passing Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.21

35 Inter-Layer Interactions (2) Downward information passing Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.21

36 Inter-Layer Interactions (2) Up-and-down information passing Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.21

37 Inter-Layer Interactions (2) Silo-wide optimization/calibration Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.21

38 Design for Change Architecture does not dictate services to be implemented Provide mechanisms for: introduce new services compose services into silos Ontology of services: describes service semantics function, data/control interfaces relationship among services relative ordering constraints Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.22

39 Ontology Networking Knowledge s 2 service s 1 m m 1,1 1,2 s 5 s m 4 4,1 application s 3 m m 1,1 3,1 m 1,2 m 2,1 m control agent 3,2 m2,2 m m 4,1 3,1 s 6 m 3,1 m 5,2 m 6,3 silos policies m m 6,1 6,1 physical layer method Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.23

40 Service Composition Constraints on composing services A and B: A requires B A forbids B A must be above (below) B A must be immediately above (below) B Negations, AND, OR Minimal set: Requires, Above, ImmAbove, NotImmAbove All pairwise condition sets realizable Forbids = (A above B) AND (B above A) Above = NOT Below Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.24

41 Service Composition Problem Given: a set of essential services application Obtain a valid ordering of these and additional services or, identify conflicts with constraints Simple composition algorithm implemented Ongoing research in formalizing the problem Input Output Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.25

42 The SILO Hourglass Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.26

43 The SILO Hourglass Applications SILO Universe SILO Transport technologies SONET OTN PPP Ethernet Physical interfaces Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.26

44 SILO Software Prototype Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.27

45 Prototype Architecture Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.28

46 SILO As a Research Tool Deploys in a slice Researcher brings: custom services tuning algorithms ontology updates Connect to measurement framework cross-layer protocol experimentation tool Con gure speci c measurement capabilities Substrate w/ programmable measurement devices Control the substrate Provide information on substrate measurement capabilities Integrated Measurement Framework Acquire measurements GENI Control Framework Sliver substrate measurement capabilities Export uni ed measurement interface Request speci c measurements Moderate access to the slice Act on behalf of the experimenter SILO Provide toolkit for cross-layer experimentation Perform experiment in a slice E EXPERIMENTER Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.29

47 Summary Vision enable flexibility, evolution: design for change fine-grain, reusable services, explicit control interface enables experimentation, flexibility, community of innovation per-flow service composition (silos) ease of evolution, policies Framework provide architectural support to vision: constrained composition commoditize cross-layer interaction / optimization Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.30

48 Ongoing Efforts New research directions silos in the core software defined optics virtualization and slicing Extend the prototype portfolio of reusable services optical testbed deployment breakable experimental net (BEN) Explore synergies with other (FIND) projects Net SILOs: A Network Architecture for Advanced Cross-Layer Experimentation NGI 2009, July 3, 2009 p.31