330 likes | 439 Views
GENI Architecture: Transition. Facility Architecture Working Group. July 18, 2007. Outline. Requirements Top Level Derived Facility Design Hardware Configuration Software Configuration Minimal Core Construction Plan. Top-Level Requirements. Generality Minimal Constraints
E N D
GENI Architecture: Transition Facility Architecture Working Group July 18, 2007
Outline • Requirements • Top Level • Derived • Facility Design • Hardware Configuration • Software Configuration • Minimal Core • Construction Plan
Top-Level Requirements • Generality • Minimal Constraints • allow new packet formats, new functionality, new paradigms,… • allow freedom to experiment across the range of architectural issues (e.g., security, management,...) • Representative Technology • include a diverse and representative collection of networking technologies, since any future Internet must work across all of them, and the challenges/opportunities they bring • Slicability • support many experiments in parallel • isolate experiments from each other, yet allow experiments to compose their experiments to build more complex systems
Top-Level Req (cont) • Fidelity • Device Level • expose the right level of abstraction, giving the experimenter the freedom to reinvent above that level, while not forcing him or her to start from scratch (i.e., reinvent everything below that level) • these abstractions must faithfully emulate their real world equivalent (e.g., expose queues, not mask failures) • Network Level • arrange the nodes into a representative topology and/or distribute the nodes across a physical space in a realistic way • scale to a representative size • expose the right network-wide abstractions (e.g., circuits, lightpaths) • GENI-Wide • end-to-end topology and relative performance • economic factors (e.g., relative costs, peering)
Top-Level Req (cont) • Real Users • allow real users to access real content using real applications • Experiment Support • Ease-of-Use • provide tools and services that make the barrier-to-entry for using GENI as low as possible (e.g., a single PI and one grad student ought to be able to use GENI) • Observability • make it possible to observe and measure relevant activity
Top-Level Req (cont) • Sustainability • Extensible • accommodate network technologies contributed by various partners • accommodate new technologies that are likely to emerge in next several years • support technology roll-over without disruption • Operational Costs • the community should be able to continue to use and support the facility long after construction is complete
Facility Architecture User Services • name space for users, slices, & components • set of interfaces (“plug in” new components) • support for federation (“plug in” new partners) GMC Physical Substrate
Generality Minimal Constraints All devices are programmable with open interfaces All devices expose multiple levels of abstraction, with the lowest level coming as close to the raw hardware as possible: sockets virtual links point-to-point multipoint virtual radio virtual wire framed TDM electrical unframed TDM electrical unframed raw wavelength Derived Requirements
Generality Representative Technology Architected to accommodate a wide range of technologies Initially selected technologies: national fiber backbone connected edge clusters mobile client devices & sensors 802.11 access network WiMAX access network cognitive radio access net Derived Req (cont)
Slicability Virtualize resources Partition resources (time & space) Isolate and contain slices Support slice composition Minimum capacity 1000-10000 total projects 100-1000 active experiments 10-100 continuously running services 1-10 high-performance/low-jitter nets Derived Req (cont)
Fidelity Node Level Network Level GENI-Wide For each selected technology (1.B.ii) support interfaces as appropriate (1.A.ii) sliver behavior (e.g., jitter) node capacity transmission capacity Derived Req (cont) • For each selected technology (1.B.ii) • points-of-presence • distribution/topology • Rich topology • Federation of autonomous domains
Real Users Wide coverage Opt-in mechanisms Generic proxies GENI-aware applications Legacy Internet connectivity number of edge connection points number of peers capacity per peer Derived Req (cont)
Experiment Support Ease-of-Use Observability Rich set of user services slice embedding experiment management legacy Internet building block services Derived Req (cont) • For each selected technology (1.B.ii) • instrumentation sensors • Data collection/archiving/analysis tools
Sustainability Extensible Operational Costs Same as 1.B.i (representative technology) Support federated ownership Derived Req (cont) • Renewal costs contribute to 1.B.ii. • Adopt sustainable software engineering practices • Architect for security • Develop monitoring & diagnosis tools
+ ISP Peers MAE-West MAE-East
Sensor Network backbone wavelength(s) Programmable Edge Node Programmable Core Node Edge Site Programmable Edge Node Programmable Edge Cluster Programmable Wireless Nodes Commodity Internet Internet Wireless Subnet
Site A PEC Internet Urban Grid Access Network Site B PWN PEN PCN GGW GGW PCN PCN PEN GENI Backbone Sensor Net PEN PWN Suburban Hybrid Access Network PWN: Programmable Wireless Node PEN: Programmable Edge Node PEC: Programmable Edge Cluster PCN: Programmable Core Node GGW: GENI Gateway PWN
Substrate Hardware Substrate HW Substrate HW Substrate HW
Virtualization Software Virtualization SW Virtualization SW Virtualization SW Substrate HW Substrate HW Substrate HW
CM Virtualization SW Substrate HW Components CM CM Virtualization SW Virtualization SW Substrate HW Substrate HW
CM CM Virtualization SW Virtualization SW Substrate HW Substrate HW Aggregates Aggregate (Proxy for set of components) Resource Controller Auditing Archive O & M Control Slice Coordination CM Virtualization SW Substrate HW
User Portals Researcher Portal (Service Front-End)
PWN PWN PWN PWN PEN PCN Wireless Subnet PCN PCN GENI Backbone PEN Wireless Subnet O&M Control Backbone Mgmt Aggregate Slice Control Operations Portal O&M Control Wireless Mgmt Aggregate Ops Team Slice Control Researcher Portal O&M Control Wireless Mgmt Aggregate Researchers Slice Control O&M Control PEC (Site 1) PEC (Site n) Edge Site Mgmt Aggregate Internet Slice Control . . .
Hour-Glass Revisited User Services Bootstrap Structure Universal Minimal Core Fixed Point Reference Implementations Substrate Components
Minimal Core • Principals • Slice Authorities (SA) • Management Authorities (MA) • User (experimenter, not “end user”) • Objects • Slices • Registered, Embedded, Accessed • Components • Data Types • GENI Global Identifiers (GGID) • Tickets (credentials issued by component MA) • rspec: resource specification • Slice Credentials (express live-ness, issued by SA)
Core (cont) • Default Name Registries • Slice Registry (e.g., geni.us.princeton.codeen) • Component Registry (e.g., geni.us.backbone.nyc) • Component Interface • Get/Split/Redeem Tickets • Control Slices • Query Status
Construction Plan • Objectives • Allow broader community to contribute (not just subs) • Scale (federate) the integration effort • Strategy • Feature Development • Roughly equivalent to open source development process • Component/Aggregate Integration • Roughly equivalent to preparing a Linux distribution
Features (hardware and software) are developed through the working group and tested locally • Once complete a feature is passed to a feature repository where acceptance testing occurs (queued until dependencies resolve) • Features in the repository can be picked up to enable development of other features Design Develop Unit Test Distribution Dependency Acceptance Testing Feature Repository Node Integration Node Testing Distribution • A collection of hardware and software features are integrated into a canonical node • The canonical node is distributed to a node repository for acceptance testing • Nodes are available for specialization Specialization Acceptance Testing Component Repository Aggregate Integration Aggregate Testing Distribution • A collection of nodes and communication features are integrated into a coherent aggregate • Aggregates are themselves integrated to create, ultimately, the GENI Facility Specialization Acceptance Testing Aggregate Repository DEPLOY