MobileMAN Integration and Experimentation of Legacy Mobile Multiho

p Ad Hoc Networks

Eleonora Borgia, Marco Conti, and Franca Delmastro

Institute for Informatics and Telematics, National Research Council

IEEE Communications Magazine July 2006 Page(s):74 – 79

報告 :羅世豪

Outline

.壹 Introduction

.貳 MobileMAN Prototype

.參 Experimental Results (Small-Scale Testbed)

.肆 Experimental Results (Medium-Scale Testbed)

.伍 Conclusions

Introduction (1/4)

Mobile ad hoc networks have not yet affected our way of using wireless networks. But now only be applied in

Battlefield scenarios Specialized civilian applications (disaster recovery, planetary ex

ploration, etc).

Users seldom use multihop ad hoc networks. This is due to three reasons

Applications (lack of) Systems implementation and integration (lack of) Experimentation (lack of)

Introduction (2/4)

So far, most of the research activities focus on The solutions of single lower-layer networking proble

ms (e.g., medium access control, topology control, routing and forwarding, transport protocols, etc.), while they almost completed neglected the upper layers.

Very large network scenarios (up to 1000 nodes and long multihop paths) where nodes move according to unrealistic mobility models.

Introduction (3/4)

MobileMAN project aimed at Developing novel solutions for open research proble

ms Integrating them with existing algorithms and softwar

e, thus contributing to tackle the lack of integration, implementations, and experimentation in the mobile ad hoc networks’ research.

Introduction (4/4)

A novel aspect of the study is its focus on the entire system rather than on isolated protocols.

From the Wi-Fi ad hoc network up to peer-to-peer (p2p) middleware services that support distributed applications such as messaging and file and content sharing.

The design goal is performance optimization for the overall protocol stack rather than optimizing the performance of a single protocol.

MobileMAN Prototype (1/10)

Testbed: small (up to 8–10 nodes) and medium-size (up to 20 nodes)

Built by following a legacy TCP/IP architecture, where the TCP/IP protocol stack runs on top of 802.11b ad hoc networks.

AODV and OLSR protocols Middleware: p2p systems on top of ad hoc netw

orks Pastry

MobileMAN Prototype (2/10)

Proactive protocols like OLSR Require periodic route updates to keep information current and

consistent, causing unnecessary routing overhead. Provide better QoS than on-demand protocols as routes to every

destination are always available and up-to-date, and hence end-to-end delay can be minimized.

On-demand protocols like AODV It establishes a route to a destination only on demand. The latency in route discovery might have a negative impact on

applications.

MobileMAN Prototype (3/10)

Currently in Mobile ad hoc systems developed Not having a middleware, but rather rely on each appl

ication to handle all the services it needs. This constitutes a major complexity/inefficiency in th

e development of ad hoc networks’ applications

In MobileMAN Prototype Have middleware platforms, by building on top of the

raw network services’ high-level mechanisms that ease the development and deployment of applications.

MobileMAN Prototype (4/10)

Ad hoc networks share concepts, such as distribution and cooperation, with p2p systems.

p2p constitutes a natural computing model for ad hoc networks as well.

Integrating p2p systems on top of ad hoc networks

Makes the variety of p2p applications and services available to mobile ad hoc network users also

MobileMAN Prototype (5/10)

Pastry works as a substrate for distributed p2p applications, offering a distributed hash table in which items can be located in a bounded number of routing hops, using a small per-node routing table.

Unique pseudo-random identifier (node Id) Each Pastry node use node ID to determine its position in a key

space, which has a ring structure. Messages are routed inside the overlay through a

subject-based routing, which exploits the key value associate to the message, and they are eventually delivered to the node with the closest identifier to the key value.

MobileMAN Prototype (6/10)

Pastry Data Structures Leafset: the set of logical neighbors of the local node.

It represents a subset of the ring centered in the local nodeID.

Neighborhood set: a subset of the physical neighbors of the local node.

Routing table: the table used to define the next hop on the ring for a given message. Each row contains a set of IDs that share with the local nodeID a prefix long as the index of the row.

MobileMAN Prototype (7/10)

:new node

:physical neighbor node

: the node logicallyclosest to the newnode

MobileMAN Prototype (8/10)

P

L

N

I’m newNeighborhood

setLeafset

N

P

L

MobileMAN Prototype (9/10)

To maintain the overlay structure Discover neighbors’ status: UDP connections Routing table data exchange: TCP connections

Introduce a high overhead in ad hoc networks, mainly when topology updates are frequent.

MobileMAN Prototype (10/10)

A small-scale testbed is a more controllable environment and it enables easier replication of the same type of experiments several times.

In a medium-scale network, the complexity of the experiments and the high variability of the topology make it almost impossible to replicate trials in the same conditions.

Therefore, the approach is Small-scale testbed: detailed analysis of the problem Medium-scale testbed: validate small-scale results, and investiga

te system scalability problems

8 nodes As routers: node B and G OLSR and AODV FreePastry Application: Distributed Messaging (DM)

Small-Scale Testbed (1/5)

Small-Scale Testbed (2/5)

Due to the limited number of peers participating in the Pastry ring, rare generations of multihop subject-based routing procedures.

Operations necessary to create and maintain the overlay introduce a significant overhead on the underlying ad hoc network, and the routing protocols’ behavior influences these operations.

Small-Scale Testbed (3/5)

Node G measures a low traffic load since it is rarely involved in messages forwarding due to its external position. For this reason, it measures essentially the overhead introduced by the routing protocol.

Small-Scale Testbed (4/5)

Since the bootstrap procedure of Pastry requires an incremental formation of the overlay collecting information from other participant nodes, the remote connections needed to create and maintain the ring generate evident traffic peaks.

Small-Scale Testbed (5/5)

A high number of TCP retransmissions and connection failures occurred, mainly due to the delay introduced to discover a route towards a destination, and to the use of unidirectional links as valid routes.

Medium-Scale Testbed (1/4)

23 nodes OLSR and AODV FreePastry Application: DM

Medium-Scale Testbed (2/4)

ICMP traffic packets loss

OLSR

Hops Packets Lose

2 15%

5 45%

6 Exceed 50%

7 67%

AODV

Hops Packets Lose

2~3 Exceed 50%

6 85%

Medium-Scale Testbed (3/4)

Average traffic load introduced by both routing protocols falls in the range of [200, 700] bytes/s.

OLSR introduces a higher overhead in the starting phase to collect the information on the whole network topology; but after this phase it soon approaches an almost constant overhead of about 250 bytes/s.

On the other hand, with AODV we observed traffic peaks when Pastry establishes the remote connections needed by the ring management operations.

Medium-Scale Testbed (4/4)

Analysis of the QoS available at the application level points out significant performance problems.

Conclusions (1/3)

Routing and forwarding performance problems By analyzing routing and forwarding operations in isolation, it

is a common understanding to consider that on-demand reactive protocols are more efficient than proactive ones, since routes are only established when needed.

However the performance with AODV are worse than OLSR due to the reactive nature of the protocol. AODV path-discovery delays have a negative impact on upper layer operations.

These delays often caused timers expiration in upper-layer protocols which, as a consequence, declared failed an operation that was indeed only delayed due to AODV delays.

Conclusions (2/3)

From the overhead standpoint As expected, OLSR produces higher routing traffic w

ith respect to AODV, but at least in the network we analyzed, the percentage of this traffic is small compared to the 802.11b available bandwidth.

Conclusions (3/3)

At the middleware layer FreePastry implementation, by operating its own rout

ing ring independently of the underlying ad hoc network, introduces a heavy overhead.

FreePastry maintenance operations exploit the TCP services and hence poor TCP performance (in mobile ad hoc networks) coupled with the FreePastry overhead reduces the overall system performance.