1. A High-Throughput Path Metric for Multi-Hop Wireless Routing
Douglas Couto, Daniel Aguayo, John Bicket, Robert Morris
MIT
MOBICOM 03

2. Problem
Min hop-count metric does not choose highest-throughput path available
Assumes links either work or don’t work
Maximize the loss ratio of each hop
Arbitrarily chooses among same length paths
Lossy links hidden by link-layer retransmission
Thresholds to discard lossy links may disconnect network
Question – Is there a “better” metric ?

3. Understanding min-hop metric Testbed
29 PCs with b radios (fixed transmit power) in ‘ad hoc’ mode
2nd floor
3rd floor
5th floor
4th floor
6th floor

4. Understanding min-hop metric Comparison against max-thruput path
better
Single hop max thruput – 450 pkt/sec

5. Min-hop Isn’t The Best
Throughput of Available Routes Between A Pair of Nodes (23 & 36)
Shortest route isn’t the best
Routes of the same # of hops provide very different throughput

6. Understanding min-hop metric Link delivery ratios at different powers
Forward direction
Reverse direction
Links with intermediate delivery ratio

7. Understanding min-hop metric Asymmetric links
At 1mW power, 28 out of 124 links have
| delivery_fwd – delivery_rev | > 25%
DATA + link-layer ACKs require both directions to work well
Inter-hop interference
Throughput = 1
Throughput = 1/2
Throughput = 1/3
Throughput > 1/4

8. ETX metric Design goals
Find high-throughput paths
Account for lossy links
Account for asymmetric links
Account for inter-link interference
Independent of network load (don’t incorporate congestion)

9. ETX metric Definition
ETX – predicted # of data tx required to successfully send a packet over link/path
ETX (link) = 1 / df x dr
ETX (path) = ∑ ETX(link)
ETX (link) measured by broadcasting periodic probe packets
Reverse-delivery ratio piggybacked in forward probe packets

10. Alternate metrics (related papers) Goal – Working with lossy channel/links
Cut-off threshold
Disconnected network
Bottleneck link (highest-loss-ratio link)
Losses accumulate
Product of link delivery ratio along path
Does not account for inter-hop interference
End-to-end delay
Depends on interface queue lengths

11. Alternate metrics (related papers) Goal - Load-balancing/congestion avoidance
Packet delay
# of paths thru node (& its neighbors)
# of packets in interface queue of node (& its neighbors)

12. ETX Evaluation Comparison with min-hop + best-route (DSDV)
better
DSDV hop-count
Best static
route found
experimentally
DSDV ETX

13. ETX Evaluation Comparison with min-hop (higher power)
DSDV hop-count
DSDV ETX

14. ETX Evaluation Comparison with min-hop+ best-route (DSR)
DSR hop-count
Best static
route found
experimentally
DSR ETX

15. ETX caveats “Broadcast” has lower priority Probe size ≠ Data/Ack size
Under-estimates data loss ratios,
over-estimates ACK loss ratios
For >= 4 hops, may choose a slower path with fewer hops
ETX assumes all links run at one bit-rate
Link-layer feedback already does a good job for DSR (& DSDV ?)

16. Conclusion / Thoughts
Proposed new metric to accommodate lossy/asymmetric links
Detailed experiments on real testbed
Favors shorter paths!
Inter-hop interference accounting problematic
Congestion / Link loss separation ?