1. FLIP : Flexible Interconnection Protocol Ignacio Solis Katia Obraczka

2. Talk overview ● FLIP overview ● Why FLIP? Motivation ● FLIP headers ● FLIP packets ● Comparison with IP ● Comparison with Diffusion ● Conclusions

3. What is FLIP? ● FLIP is a network protocol that aims to be flexible. It tries to reduce the overhead as much as possible for small devices but does not limit the functionality of more powerful ones. ● Configurable by higher layers (Header morphing)

4. Why FLIP? ● Generic protocols have too much overhead for small devices. ● Specific protocols are not general enough. ● Applications need access to the lower layers to optimize use. ● Every bit counts

5. Sensor Networks ● Data gathering ● Small power constrained devices ● Wireless communication ● Long Lifetime ● Large Scale ● Specific Tasks ● Unattended

6. What about IP? ● Overhead ● Addressing scheme? ● Routing? ● Fragmentation? ● Size? ● etc.

7. Fields defined by FLIP ● Version (1 byte) ● Destination (2, 4 or 16 bytes) ● Source (2, 4 or 16 bytes) ● Length (2 bytes) ● Time To Live (1 byte) ● Flow (4 bytes) ● Protocol (1 byte) ● Checksum (2 bytes)

8. The Meta-Header Bitmap ● The meta-header bit map defines which fields will be included in the header. ● Each header field will be represented by one or more bits in the meta-header. ● If the bit is on, the field will appear.

9. The continuation bit ● We don't really need the whole meta-header bitmap since not all fields might be required. ● The bitmap is divided in groups which are then placed on different bytes.

10. The ESP packet ● Extra-Small-Packet ● For special very small payloads (6 or 14 bits)

11. 11 isolis@ucsc Current Work on FLIP FLIP HeaderGTP Header GTP flags

12. 12 isolis@ucsc Sample FLIP Packets FLIP ESP packet (extra simple packet) FLIP/GTP packet

13. Sample API ● Uses standard socket interface

14. 14 isolis@ucsc FLIP Functions

15. Comparison with IP The special cases of Destination and Source and Destination only use 2 byte addresses. Percentages are overhead of header compared to data. Packet sizes for 1 and 1000 byte payloads

18. 18 isolis@ucsc Directed Diffusion ● Sink node collects data ● Sink floods an interest to the whole network, establishing reverse paths to itself. ● Nodes that have the data the sink is interested in report back (multiple-paths). ● Sink reinforces the best path, that is, requests a higher data rate on that path. ● The interest is flooded periodically ● Newer versions of Directed Diffusion have other mechanisms

19. 19 isolis@ucsc Scenario 1: Diffusion Packets Diffusion Static Header Flexible Header Can we optimize Diffusion by using FLIP?

20. 20 isolis@ucsc Diffusion Monitoring

21. 21 isolis@ucsc Scenario 2: Simple Data Gathering

22. 22 isolis@ucsc Simulation Parameters ● NS-2, 2000m x 2000m area, 10 runs, 21 secs. ● 300 nodes, no movement. ● 250m transmission range. ● 802.11-like MAC. 660 mW / 395 mW ● Starting energy is 1 Joule ● Interests are every 5 seconds ● 1 source, 1 sink requesting 10pkts/sec

23. 23 isolis@ucsc Simple Data Gathering

24. 24 isolis@ucsc Adding Data Aggregation ● Aggregate data as it flows through the network ● Data must meet certain criteria ● Not all data can be aggregated ● Lossy & lossless aggregation

25. 25 isolis@ucsc Scenario 3: Ring Aggregation

26. Conclusions ● FLIP incurs in small overhead when providing IP functionality. ● Header overhead on special cases can be very small. For example on very small payloads. ● It does not try to replace protocols such as IP. ● More research is needed since many variables are yet to be determined.

27. Thank You Ignacio Solis isolis@cse.ucsc.edu http://www.cse.ucsc.edu/~isolis/