opensourcejason.info
mobi share session 1
CodeTorrent
- Multimedia-based proximity marketing
- Movie trailers in nearby theatres
- On VANET
- Based on BitTorrent file swarming from same group called CarTorrent
- Noted limitation: Missing coupon problem
- Use network coding to solve problem
- Basic CarTorrent Idea:
- Single-hop puling instead of CarTorrent multihop
- Cars exchange randomly combined packets
- They share our belief on single-hop vs. multi-hop routing in ad-hoc (i.e., single hop is better)
- Networking coding
- Matigate rare pice problem
- maximize benefits of overhearing
- Periodic beaconing for discovering helpful nodes (who has at least one LI encoded block)
- Simulation in QualNet using real-track mobility model
- CodeTorrent
- scales to mobility
- favors cooperations
- eliminate missing coupon problem
- Future work
- model impact of mobility
- real testbed.
XScribe
- P2P can be a booster for MANET deployment
- Efficient P2P services are a more friendly environment (zero-conf)
- Designed around opposite assumption
- P2P: scale to thousands of users
- resource-rich environemnt
- trade consumption for scalability
- MANET
- hundreds of nodes are unreasonable
- resource-scarce
- rade everything for reduce bandwidth consumption
- P2P over MANET
- legacy typically fails, but it's feasible, and worth it
- success: CrossROAD, VRR
- Legacy P2P: Scribe
- multicast protocol working on top of a Pastry-like DHT
- straightforward to use structured overlays for mcast
- build shared tree based on a standard reverse-path algorithm
- implemented on top of overlay, subject-based routing identifies root and builds tree
- self-organizing, self-healing, decentralized
- scribe data delivery: get everything to the root
- cons
- centralized
- several hops in the overlay
- XScribe
- tries to address centralization/structure issues (avoid synchronization)
- avoid serveral-hop problem
- pure p2p (overlapping path exploitation) not addressed
- make all nodes in the group aware of receiver set
- send send a distinct mesage to each receiver
- Used core ideas of CrossRoad (pastry-based DHT) to make receiver awareness efficient
- embed group membership into L3 routing ads
- no extra MAC-level access
- CrossROAD makes all nodes just one-hop away from each other
- Experiments
- Collaborative whiteboard application
- packet loss halved on most nodes (xscribe vs. scribe)
- slight increase on some nodes
- significant delay decrease
- Conclusion
- solve Scribe problems of centralization and structure-induced path stretch by efficient dissemination of group membership and using CrossROAD-based ideas to avoid path stretch
- Critique
- Trend of performance worsening with load is not fixed
- Lesson learned
- pure p2p delivery is not affordable in MANET
- p2p mcast has to meet L3 mcast
- Audience question:
- Xscribe seems to be a one-hop overlay based on OLSR, but there is already mcast work in OLSR, why is it not compared.
- What about scalability?
- Response: goal 20-30 to nodes because of flat-ad-hoc limitation
- Followup: Is this scheme worth it with such a small network?
GUESS
- Assumed biggest limitation: spectrum availability
- Shows typical FCC allocation image
- Observation: paid-for spectrum is unused in many areas
- Opportunistic Spectrum Sharing: OSS
- Already legislatd for a small section of 5Ghz
- proposed for VHF/UHF
- Requires smart sensing radios that dynamically tune
- This paper focuses on sensing available medium
- Observation: Primary user detection is non-trivial (shadowing)
- Solve it via coordination approach between secondary users
- Requirements:
- Large amount of sensory data
- Short timescale (need quick decisions)
- Minimal overhead
- First-order approach:
- exchange complete info with all devices (structure-based)
- Communicate periodically
- not robust to failure and change
- may be excessive
- Spectrum Sensing Obersation
- summaries are sufficint for coordination
- compute statistics for a band
- Idea 1: In-network aggregation
- spectrum usage changes incremenationally
- no need for periodic recompute, energy changes slightly over time
- Ideas 2: We need update protocl
- Contributions
- Novel application of randomized gossip to spc sensing
- Apply probablistic F-M algorithm for aggregation
- Extend system to support incremental updates
- Aggregation issues
- duplicate messages
- large message size
- Solution:
- F-M bit-vector approach (probabilistic summaries, sorta like lossy compression)
- F-M COUNT extended to SUM by varying "coin toss" depending on signal strength (more for higher)
- roblem: F-M aggregate can not be updated
- Basic sol: maintain an addition F-M delete vector for descrease in value
- Finally: Incremental Random Walk Info exchange protocol
- Only initiate random walks from nodes with new info
- Inc random walk converges up to almost 3 times faster in simulation
- See 2.5x reduction in overhead
- Future work:
- limit impact on distant nodes
- Avoid minor updates
- Implentation considerations (DNE yet)
- Audience Question:
- Why is this important over 1000 nodes
- answer: that's the extreme density case
- Question: can you also use long term statistics?
- Answer: yes, for fixed infrastructure/environment
- Question: how do you handle biasing because you use bit aggregate
- Answer: some things addressed in the paper (multiple