Routing and Transport challenges in mobility-assisted communication

Konstantinos Psounis

Assistant Professor

EE and CS departments, USC

2

Intermittent connectivity lack of contemporaneous end-to-end paths

Disaster communication Vehicular ad hoc networks Sensor networks for environmental

monitoring and wildlife tracking Ad hoc networks for low cost

Internet provision to remote areas Inter-planetary networks Ad-hoc military networks

Routing: “store-carry-and-forward” model Transport: message-oriented approach, link-layer retransmissions Interoperability with “traditional” network segments also a goal

S

D

The need for mobility-assisted communication

3

3

Example of store and forward routing

S

D

1

2

4

5

6

7

8

9

10

11

12 13

14

1615

4

Routing

Redundant copies reduce delay Too much redundancy is wasteful and induces a lot of interference Middle ground:

spray a small number of copies to distinct nodes use carefully chosen relay-nodes to route each copy towards the destination

Challenges

How many copies to use?• derive formal expressions that take into account real world limitations and

compute number of copies that guarantee performance targets

How to optimally spray the copies• use stochastic optimization and portfolio theory to find optimal policy

How to optimally choose relays?• find a good utility function that indicates the goodness of a node as a relay

5

0

50

100

150

200

epidemicrandom-flood

utility-flood spray&waitspray&focus

Transmissions (thousands)

K = 40 (8.6%)

K = 50 (14.8%)

K = 60 (27.7%)

K = 70 (52.9%)

K = 80 (79.2%)

0

500

1000

1500

2000

2500

3000

epidemicrandom-flood

utility-floodspray&waitspray&focus

Delivery Delay (time units)

How well spraying-based routing works?500x500 grid, 200 nodes, medium traffic load

Spraying schemes outperform flooding schemes in terms of both transmissions and delay

As connectivity increases delay of spraying schemes decreases delay of other schemes increases due to severe contention

Tx Range K (connectivity: % of nodes in max cluster)

6

How many copies to use?

Number of Copies L (M = 100)

Required Percentage of Nodes (L / M) Receiving a Copy

0

2

4

6

8

10

12

14

100 1000 10000 100000

Number of Nodes (M)

percentage (%)

α = 2

α = 5

α = 10

= expected delay of spraying schemes over the expected delay of an oracle-based optimal scheme

to be within some distance from optimal

7

How to spray the copies?

lth

B closer to D

A closer to D

Practical heuristic: if l lth (a few copies)

• the best node should keep/get all copies

else (a lot of copies)• do binary spraying (split copies in

half)

Optimal policy: node A has l copies for node D node A encounters node B

150x150 grid, 40 nodes, K=20

8

Transport

Message oriented transport rather than stream-oriented (no concept of flow)

Link layer retransmissions hard to support end-to-end feedback mechanisms

Congestion control: short term relief: if a node is congested give it priority over other nodes

that contend for the same medium• challenging to identify and coordinate these nodes in practice

medium term relief: use congestion information to dynamically adapt routing paths

• e.g. lower utility of congested nodes Of course, source rate adaptation should eventually occur if network is

oversubscribed

9

Set of contending nodes

SR

Congestion control and fairness require coordination among contending nodes

Which are those nodes? assume, for simplicity, a single disk model for the transmission and

interference range

10

Interoperability

Future network: Wired core Wireless edge

• single-hop wireless sub-networks (SWN)

• multi-hop wireless sub-networks (MWN)

Use core-edge elements to break connections into sub-connections

mask differences

Delay/disruptive tolerant MWN

Mobile Ad-Hoc MWN

Sensor/Mesh MWN

Core-EdgeElement

A

Bc

B

Ac

WiFi SWN

Base station

WiMax SWN

11

Core-edge element functionality examples

Transport connection management Hide latencies and disconnections from the wired core e.g. delay the start of successive sub connections until enough data

are accumulated

Packet caching Core-edge element acts as proxy of sender or receiver e.g retransmit cached packets in case of loses

• no requirement to contact (hard to locate) source

12

Experimentation and applications

Human mote experiments students carry motes within main campus and on its vicinity

USC testbed hundreds of static nodes arranged in disconnected clusters

(tutornet platform) and a handful of radio-capable robots (robomote project) to bridge the gaps between them

Applications offer connectivity for delay tolerant applications to USC

commuters• in collaboration with the university transportation office

customize protocols for VANET applications

13

Selected Publications and funding sources more info available at http://ee.usc.edu/research/netpd/publications/

Publications: Routing

Efficient Routing in Intermittently Connected Mobile Networks: The Multi-copy Case, T. Spyropoulos, K. Psounis, and C.Raghavendra, to appear in IEEE/ACM Transactions on Networking, February 2008.

Efficient Routing in Intermittently Connected Mobile Networks: The Single-copy Case, T. Spyropoulos, K. Psounis, and C. Raghavendra, to appear in IEEE/ACM Transactions on Networking, February 2008.

Performance Analysis of Mobility-Assisted Routing, T. Spyropoulos, K. Psounis, and C. Raghavendra, ACM MOBIHOC, Florence, Italy, May 2006.

Transport Interference-aware fair rate control in wireless sensor networks S. Rangwala, R.

Gummandi, R. Govindan, and K. Psounis, ACM SIGCOMM, Pisa, Italy, September 2006. Mobility

Modeling Time-variant User Mobility in Wireless Mobile Networks, W.-j. Hsu, T. Spyropoulos, K.Psounis and A. Helmy, IEEE INFOCOM, May 2007.

Funding: External: NSF Nets Internal: Zumberge foundation, startup funds