1 / 19

Network Architecture Research in SPYCE CIP/URI Project

Network Architecture Research in SPYCE CIP/URI Project. Jonathan M. Smith University of Pennsylvania 3/31/03. Highlights. Network support of diffuse computing High-performance active networking for monitoring and measurement

Download Presentation

Network Architecture Research in SPYCE CIP/URI Project

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Network Architecture Research in SPYCE CIP/URI Project Jonathan M. Smith University of Pennsylvania 3/31/03

  2. Highlights • Network support of diffuse computing • High-performance active networking for monitoring and measurement • Ultra-large-scale peer-to-peer architecture for simulations and multiplayer games • Diffuse computing network measurement infrastructure – “cing” • Transcoding and proxies to support diffuse computing on media streams and images in large-scale heterogeneous nets

  3. The 70s Internet is no more! • New Roles for Hosts (Hosts, WWW, P2P) • Routers *plus* RFC3234 “middleboxes” • Other Diffuse Computing Elements

  4. Network Challenges • Diagnosis in spite of complex network conditions, heterogeneity • Reconfigurability for high availability • Accommodating new applications (rapidly) • Accommodating new tech rapidly (wireless) • Scalability; Metcalfe’s Law phenomenon • Need for global security even with local failures and subversions

  5. Achievements/Prototypes • LAME/FLAME* active networking • fast/safe/extensible NW monitoring • NOMS 2002 • CING* network measurement tool • characterize Internet paths (Performance’02) • Indirect measurements (Infocom’03) • Transcoding proxies* • P2P massive multiplayer games* • Under submission • Applications to distributed simulation • 3 slides on FLAME; rest on latter two systems * Done with CIP/URI support

  6. Diagnosis: AN-based monitors: Uses: IDS DDoS/virus detection performance debug traffic engineering traffic measurement Accounting SPYCE platform D D O S a c c t Traf. Eng. monitor Controlcommands Packet flow router network

  7. FLAME: Safe in-network monitoring • Hosts of varioustypes • Routers / FLAME nodes • Diffuse Computing Elements • P2P participants Router Router FLAME FLAME Router FLAME

  8. New results • FLAME network monitoring: • Role and cost of safety (Cyclone in OpenBSD) • Router integration issues (move from OpenBSD software router to 3Com Corebuilder 3500 and other commercial platforms) • Market-based resource control with AN: • Further mechanisms – focus on routing • Certified resource bounds in packet headers • BGP->S-BGP via incentives…

  9. Transcoding proxies • Client/Server model too simplistic • “remote access” style of distributed computing from 70s • Convenient model for LANs – Network File System, Remote Procedure Call, etc. • Diffuse computing allows a far more general approach • WWW can be seen as distributed RPC • Diffuse transcoding for network and device heterogeneity

  10. Transcoding Proxies: The Internet Media Server sends Packets with instructions Source=Proxy Client Limitations Meet Server Directed Transcode Destination=Proxy Wireless/ telephone The Internet “Browser” User “Device”

  11. Diffuse Computing issues addressed • Distributes work • Easy (in fact desirable) to have multiple proxies • Load-balancing with anycast • Anycast is path to economics control • Addresses scale • Copes with heterogeneous nets and devices • Localizes complexity • Diffuse complexity, not centralized at server Proxy Client Server Client Server Proxy

  12. Massively Multiplayer Games • Key area of DoD study (e.g., 2002 ISAT) • Commercial systems have outpaced DoD distributed simulations in scale and function • Many applications to constructive simulations and hybrid simulations • Diffuse Computing is the path to ultra large scale, robustness, extensibility, etc. • Natural application of P2P systems • What are the systems/networking issues?

  13. P2P for Massively Multiplayer Games • Hosts – Participate Dynamically as Peers • Routers • Participating Nodes create an overlay

  14. Game State • Highly detailed, becoming more so • Sets of objects of interest to players • Weapons • Treasures • Buildings, resources, etc. • Updates as players alter the world • Broadcast does not scale • Need consistent game state • Issue: partitioning (link/node failure) • Issue: scale (200,000+ players in some systems)

  15. Discoveries in MMG research • Interest in game state is localized • Players not interested in global picture • Also noted in DoD HLA interest management • This can be exploited! • Caches/ local replicas / define regions • New consistency protocol for state updates • Reliability with small # of replicas • Implemented in a simulator for the Pastry P2P system • Extensive simulations suggest highly scalable performance, dependent on region structure

  16. Diffuse Computing issues addressed • Ultra large scale • Focus on global reliability with local failures • Tight coupling between networking and distributed computing issues • Ongoing work, with excellent opportunities to introduce incentives for diffuse control (e.g., for dynamic region construction) • Great opportunity for DoD tech transfer

  17. MANY PLACES for Diffuse Computing! • Hosts • Routers • Network Embedded Diffuse Computing Elements • Peer-to-peer simulations and multiplayer games Cing P2P Client Proxy FLAME Proxy P2P P2P

  18. Open Questions, Future Work • Infuse economic models into proxy model • Move MMG system from emulator to Pastry (straightforward but mucho engineering…) • Provide economic incentives to increase locality (adaptive regions) and further accelerate consistency protocols • Continue to extend and integrate new systems into SPYCE platform (e.g., AN+P2P) • Embed incentives in algorithms to “coerce” users towards secure systems, e.g., S-BGP

  19. Notes: • MMG work done by Prof. Honghui Lu, Dr. Bjorn Knutsson, Xu Wai • Transcoding Proxy work done by Prof. Honghui Lu, Dr. Bjorn Knutsson, Dr. Jeffrey Mogul • Cing work done by K. Anagnostakis, Prof. Michael Greenwald, R. Ryger • FLAME work done by K. Anagnostakis, et al. • Papers + Software Distros at: http://www.cis.upenn.edu/~spyce

More Related