1 / 9

Perspective on Overlay Networks

This panel explores the challenges and implications of computing on a massive scale, including the benefits of overlay networks, infrastructure scalability, dynamic environments, extending the scope of overlays, introspective tuning, and security challenges.

reads
Download Presentation

Perspective on Overlay Networks

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. Perspective on Overlay Networks Panel: Challenges of Computingon a Massive ScaleBen Y. ZhaoFuDiCo 2002

  2. Why Overlays • Changes to standards is hard • Possible benefits of overlays • Extensible communication models(more than best-effort point to point) • More easily deployed services / infrastructures • More adaptable, fault-tolerant … ravenben@eecs.berkeley.edu

  3. Massive Scale • It’s coming… • Driven by global scale resource sharing apps • File-sharing, data dissemination, shared computation • Viral deployment (Kazaa / alt.net) • More thin clients going online: cellphones, PDAs, fridges, toasters ravenben@eecs.berkeley.edu

  4. Implications of Scale • Pushing bounds of infrastructure scalability • Logarithmic bounds are not enough! • Worse case: massive congestion, network partitions, data / service unavailability • Scrutinize infrastructure scalability • Re-examine system level primitives • Control traffic, fault-detection, replication overhead • Dampen impact of operations on infrastructure • Localize bandwidth / storage requirements, faults ravenben@eecs.berkeley.edu

  5. Highly Dynamic • It’s here… • Small half-life for P2P apps (leeching) • Connectivity loss from congestion, flash crowds, misconfiguration, hardware faults, lower MTBF • BGP convergence 3-15 mins • Mobile clients in semi-connected mode • E.g. chatting / file-sharing on your IPAQ / Palm ravenben@eecs.berkeley.edu

  6. Implications of Change • Have to reexamine stability assumptions • Overlay maintenance under instability • Without care, worst case: sub-optimal paths, network partitions, broken invariants  loss of availability • Proactive overlay construction and maintenance • Integrate optimal routes from square 1 • Estimate global knowledge w/ distributed algorithms at node insertion • Run-time heuristics are too late • Constant fault-detection and self-repair • Use of redundancy to mask self-repair ravenben@eecs.berkeley.edu

  7. Extending Scope of Overlays • Broader utility than file storage and retrieval • General routing layer (efficiency, fault-tolerance) • Detour, RAMP, RON, Tapestry • Flexible data dissemination layer • overlay multicast, anycast, mobility • Collaborative applications • Shared annotations, mail filtering • Data transcoding • Shared workspaces, games • Distributed computation • Barter incentives • (Mojos, ads, access to resources) ravenben@eecs.berkeley.edu

  8. Introspective Tuning for Reality • Exploit knowledge of the environment • Network latency discontinuity • Order of magnitude latency jump from LAN to WAN • Generic solution:Brocade: minimize wide-area hops via 2nd overlay • Non-uniformity in node resources • PDA < laptops < desktops < servers • Segregate supernodes, infrastructure routers, pure clients, storage devices ravenben@eecs.berkeley.edu

  9. Many Security Challenges • Open P2P = Infrastructure for attacks? • Misreporting data  unfair resource allocation • DDoS attacks on infrastructure • Sybil attack  enabling scale for attacker • Detecting maliciousness made harder • Scalable defense? • Cryptographic methods for authentication, privacy, etc. • Byzantine agreement • Redundancy (ID aliasing, multiple namespaces) • Misuse leads to resource based DDoS • Access control? ravenben@eecs.berkeley.edu

More Related