Overcoming the Internet Impasse through Virtualization Thomas Anderson, Larry Peterson, Scott Shenker, Jonathan Turner 원종호 (INC lab) Sep 25, 2006
Outline Introduction Three requirements Virtualization Physical testbeds & overlays Virtual testbeds Virtualization : means or ends Conclusion SNU INC lab. 2
Introduction Internet has dramatically changed from a research project to a critical and nearly ubiquitous infrastructure Current Internet architecture is ill-suited for a number of challenges But, the prospects for fundamental change appear slim. barriers to experimentation needed to evaluate new ideas barriers to deployment Architectural barnacles impair the Internet s longterm flexibility, reliability and manageability. SNU INC lab. 3
Introduction - Slow-moving ship with barnacles SNU INC lab. 4
Three requirements Researchers must be able to experiment easily with new architectures on live traffic. Plausible deployment path for putting validated architectural ideas into practice. New architectures that address broad range of architectural problems (not just narrow solutions) SNU INC lab. 5
Virtualization High-level abstraction that hides the underlying implementation details Core principle in overlays Allowing nodes to treat an overlay as if it were the native network Allowing multiple overlays to simultaneously use the same underlying overlay infrastructure. SNU INC lab. 6
Physical testbeds Production testbeds Internet2 Closely tied to current internet architecture Users have no choice about participating in the testbed Users don t realize their traffic is part of an experiment Conservative on their experiment SNU INC lab. 7
Physical testbeds Research testbeds DETER (Defense Technology Experiment Research) For cyber-security research Not carry traffic from a wide variety of real users Production & Research are not adequate for evaluate new architectures SNU INC lab. 8
Overlays Deploy narrow fixes to specific problems Performance, availability, DOS, content distribution, multicast. Each of problems isolated. Architecturally tame Emphasis is on deployment in today s Internet. Most current overlays assume IP. Thus, current overlay will likely become another barnacle. Need philosophical revolution. not technical alteration. SNU INC lab. 9
Virtual Testbed Key features Overlay substrate Composed of dedicated but multiplexed overlay nodes. Multiple experiments can run simultaneously on same infrastructure. Client-proxy mechanism. A host can opt in a particular experiment on a specific substrate overlay Above two features resolve barrier-to-entry & architectural limitations that overlay faced SNU INC lab. 10
Virtual Testbed Technology overview PlanetLab software architecture multiplexes multiple slices. Each slice runs different network service, application or architecture. SNU INC lab. 11
Virtual Testbed Technology overview PlanetLab have some technical issues Achieving sufficiently high throughput QoS Proxy SNU INC lab. 12
Proxy technology DNS request proxy True IP or fake IP VT ingress node Virtual testbed VT egress node Legacy server fake IP user True IP Normal Internet SNU INC lab. 13
Virtual Testbed Service hosting PlanetLab can host a service within the virtual testbed that remains visible to nonparticipating clients. DNS resolution to point the client to a nearby virtual testbed representative. (like CDN) The local representative translate the packets into an internal format for delivery to the server translate the packets back to Internet format for the replay SNU INC lab. 14
Virtual Testbed Inspiration Virtual testbed borrows the idea from X-bone. Dynamic Internet overlay deployment and management Its tools support automated establishment overlay NLR (National LambdaRail) High speed backbone, optical infrastructure PlanetLab + NLR Support larger traffic volumes. PlanetLab-based overlays serve as access network for the backbone Developing and deploying become easy. SNU INC lab. 15
National LambdaRail SNU INC lab. 16
Virtual Testbed Deployment Can we find a plausible deployment? 1. A NGSP chooses a particular new architecture 2. constructs an overlay supporting that architecture 3. distributes proxy software that lets anyone, anywhere access its overlay 4. NGSP users would still be purchasing Internet service from their ISP 5. But if the overlay is successful, NGSP or ISP would offer direct access to customers Cf. NGSP = New-generation service provider SNU INC lab. 17
Virtualization means or ends Virtualization techniques are used for experiment and deployment. Virtualization can play a central role If new architectures always compete against the old. If many narrowly targeted architectures exist simultaneously. Supporting overlays will be core functionality of the architecture. Virtual link establishment Proxy-like reachability SNU INC lab. 18
Virtualization means or ends Redefining Internet architecture - Purist vs. Pluralist Purist A single universal protocol Overlay is necessary evil Virtualization is not a fundamental aspect of the architecture just means to install new architecture. Pluralist IP is only one component Overlay offer one more way to deliver the service users want Dynamic and evolving architecture can be defined as the union of the various overlays and protocols. A pure architecture for the high-speed core A more pluralist architecture closer to the edge SNU INC lab. 19
VINI Virtual network infrastructure (VINI) allows network researcher to evaluate their protocols and services in a realistic environment. Provide a high degree of control over network conditions Allow researchers to deploy and evaluate their ideas with real routing software, traffic loads and network events Supports simultaneous experiments with arbitrary network topologies on a shared physical infrastructure. SNU INC lab. 20
Conclusion Non-incremental architectural change has little chance of adoption. traditional testbeds are no longer an effective. Thus, research community has narrowed its focus. Empirical and incremental research are not sufficient to meet fundamental challenges the Internet faces. Providing easy access to virtual testbeds, there will be a renaissance in applied architectural research. SNU INC lab. 21