29
Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua University

Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

Embed Size (px)

Citation preview

Page 1: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

Scalable Routing in Delay Tolerant Mobile

Networks

Hao Wen1

Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu1 Department of Computer Science, Tsinghua University

Page 2: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 22/51/51

Outline

Background

Related work

Region-based mobility pattern

Protocol design

Evaluation

Conclusion

Page 3: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 33/51/51

Background

As a kind of challenged networks, Intermittently Connected Mobile Network (ICMN) is a Delay-Tolerant Mobile Network (DTMN) that is made up of mobile nodes

Intermittent Connectivity

Variable Delay

Key metrics in DTMN:Packet Arrival Rate

Delay

Scalability (not considered by most previous work)

Page 4: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 44/51/51

Typical cases of DTMNPrinceton ZebraNet: Track and monitor Zebra in Africa

UMass DieselNet buses: VANET

Background

Page 5: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 55/51/51

Big challenges for routing protocols in DTMNDue to frequent network disruption, it is difficult to establish reliable end-to-end paths between mobile peers.

Carry-and-forward is being widely used. However, it will bring much overhead compared with traditional routing.

The constraints brought by mobility and resources make the routing problem much more challenging, especially for resource-constrained devices such as sensor nodes or Bluetooth devices.

Background

Page 6: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 66/51/51

Background

In this paper, we take advantage of the spatial property and propose two scalable protocols based on regional movement,

Since the long-term spatial property is relatively stable over time, our protocols avoid complicated computation for delivery probabilities and excessive storage for tracking encounter history.

Page 7: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 77/51/51

Outline

Background

Related work

Protocol design

Evaluation

Conclusion

Page 8: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 88/51/51

Flooding routing is extremely wasteful of limited resources, such as wireless bandwidth and storage space.Distributing a bounded number of copies to reduce the overhead [Spray and Wait, WDTN 05]Replicating packets with a small probability [Wireless Network 2002]They do not make use of gaining knowledge about network conditions, so their performance is not satisfying under more realistic conditions.

Related Work-Flooding Type

Page 9: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 99/51/51

Based on encounter history, they estimate delivery probabilities between nodes.

RAPID [Sigcomm 07]Translates the routing metric into per-packet utilities

Calculate utility according to average meeting time

Every node records meeting history of all nodes

This type exploits past knowledge of encounters but faces a challenge of choosing the right time scale :

Short scale: “distance effect”, the long-run trends are difficult to capture from a short-scale temporal way.

Long scale: consuming much storage and computation

Related Work- Utility Type

Page 10: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 1010/51/51

There are many recent works on WLAN measurements which reveal the important spatial properties of the real-world users.

They imply that not only temporal but also spatial info can be taken into account during the design of practical routing protocols.

Related Work

Page 11: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 1111/51/51

Outline

Background

Related work

Protocol design

Evaluation

Conclusion

Page 12: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 1212/51/51

Our solution is motivated from a simple observation: location-preference and re-appearance are usually observed as typical features not only from human beings but also from other species:

In a short time, node may move in somewhat random way or paused at some location

In a long-time scale, node change position among areas that related to their lifestyle.

From a social context, visited locations and people's encounters both have a strong connection with the affiliation and lifestyle. Different from encounter-based protocols, space-based pattern offers us a different view.

Protocol design

Page 13: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 1313/51/51

The main idea of our space-based protocols is to find the popular regions for destination node and distribute copies of packets inside those popular regions.

The concept of space is defined as a kind of logic converge

We simply adopt a square coverage as a unit space and assume nodes are equipped with any localization method.

Protocol design

Page 14: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 1414/51/51

Step 1: calculate distribution probability matrix R

Step 2: choose candidate destination regions based on two methods

Step 3: the source node will send one copy to every chosen region with the help of relay nodes

Step 4: after the relay node reaches the destination regions, it will trigger a spray-and-search distribution

Protocol design

Page 15: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 1515/51/51

Every unit time, every node will track the current region and record into space transition matrix B. Then we can get distribution probability matrix R and transition probability matrix P

Protocol design-Step 1

Page 16: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 1616/51/51

Protocol design-Step 2

In terms of choosing candidate destination regions, we propose two space-based protocols according to probability or distance, respectively.SpacE-Probability (SEP) optimized protocol

regions are simply chosen according to the distribution probability rj(y) of node y:

Given the probability threshold P, L regions are chosen in decreasing order of rj(y)

Page 17: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 1717/51/51

Protocol design-Step 2

SpacE-Distance (SED) optimized protocolThe source node x makes decision based on the Euclidean distance dkj from the current region k to the destination region j. The problem is formulated to

To reduce the complexity of this 0/1 Knapsack Problem we use a greedy algorithm: choose L regions in decreasing order of distance value per unit of probability weight, i.e.,

Page 18: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 1818/51/51

To forward copies to L chosen regions, relay nodes are chosen according to the distribution probability in SEP or the expected distance in SED.

SpacE-Probability (SEP) optimized protocolWhen two nodes encounter in the region v (v!= j), the copy will be forwarded to the node that has a bigger rj .

SpacE-Distance (SED) optimized protocolNodes will exchange transition probability pvi and calculate expected distance from v to j. The copy will be forwarded to the node that has a smaller Dvj.

Protocol design-Step 3

Page 19: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 1919/51/51

When relay node Z arrives at destination region j, choose w nodes to take copies using Spray and Search

Spray phase: Z distribute data using binary forward to w nodes (”jump-start” spreading in a quick manner )

Search phase: the copy will be forwarded to a better relay according to the policy in step 3.

Protocol design-Step 4

Z

Z

Z

Page 20: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 2020/51/51

Outline

Background

Related work

Region-based mobility pattern

Protocol design

Evaluation

Conclusion

Page 21: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 2121/51/51

Simulator: Opportunistic Network Environment (ONE) [Jorg Ott, 2008]

Epidemic: a greedy strategy

Spray and Wait: a passive strategy

Prophet: based on the encounter frequency

Maxprop: based on the last encounter time

The numbers of copies of SEP/SED and SNW are both restricted to 10% of all n nodes.

The overhead is defined as

Evaluation-Experiment setups

Page 22: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 2222/51/51

Bus Network Model [Jorg Ott, Mobility Model 2008]Based on the city area of Helsinki with 10000 × 10000 m2.

Total n buses are evenly distributed in 8 bus routes

The buses move at 7-10m/s with a 10-30s waiting time at each bus stop.

RENA generates 16 regions with equal unit size.

Evaluation-Experiment setups

Page 23: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 2323/51/51

Epidemic and Prophet increase firstly and then slowly decrease to about 60%

MaxProp achieves the best performance when n < 72 and then deteriorate rapidly due to the fast increase of computation and control packets

SEP, SED and SNW gradually increase to stable values

Evaluation-Bus Network

Page 24: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 2424/51/51

Due to the estimation of delivery probability based on the encounter frequency, Prophet spent less overhead than Epidemic.

SNW consumes the least overhead as a result of passive waiting.

By calculating global shortest path based on the last encounter time, MaxProp only behaves well when the number is small.

Using limited computation and storage, our proposed protocols achieves not only better scalability but also high performance.

Evaluation-Bus Network

Page 25: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 2525/51/51

Although our protocols are proposed based on the spatial property, they could achieve feasibility in most scenarios.

In the extreme case without any location-preference property, such as random mobility models, our protocols could naturally transfer to SNW:

when there are no candidate destination regions, they simply chooses its current region as destination and starts binary spray.

So the delivery performance of SNW is the lower bound.

Evaluation-Performance Bound

Page 26: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 2626/51/51

In this paper, we take advantage of the macro-level spatial information and propose two space-based probabilistic forwarding protocols. Different from temporal-based protocols using encounter history to calculate the delivery probability, our protocols make good use of regional movement pattern hidden in spatial property and need not expend much computation for calculation. Compared with several typical protocols in a bus network, the scalability of our protocols is well verified.

Conclusion

Page 27: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 2727/51/51

Thank you for your attention.

Q&A

Page 28: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 2828/51/51

Backup

ParametersConnectivity: interpersonal communication (10m range, 2 Mbps) using Bluetooth devices.

Every t interval time, one mobile user generates one packet of 1 KB to a random destination.

Page 29: Scalable Routing in Delay Tolerant Mobile Networks Hao Wen 1 Jia Liu, Chuang Lin, Fengyuan Ren, Chuanpin Fu 1 Department of Computer Science, Tsinghua

QoS Lab in Department of Computer Science & Technology, Tsinghua UniversityPage Page 2929/51/51

Backup-Region Size

Different choices depend on the specific granularity of the movement in different applications. In addition, we should also consider the storage occupied by control packets. (the average storage per region consumed by control overhead is about 650 Bytes).

In this paper, our empirical analysis suggests that it is better to keep control packets under 20% of the whole storage.