Supporting Groupware in

Mobile Networks

Idit Keidar,

Technion – I.I.T

Joint work with N. Lavi and I. Cidon

Agenda

Motivation for mobile groupware Current solutions Our proposed architecture Group management- one solution Simulation and analysis Future work Conclusions

Current Application Trends

Groupware and collaborative applications are widely used. Chat, Instant-Messaging, VoIP, VCoIP, Net-meeting Exchange, Lotus notes, webex Multiplayer interactive games Push-to-talk (PTT)

Current Cellular Trends

Simple groupware such as Instant Messaging widely used

Major cellular providers (Orange, Verizon, Nextel) offer PTT services

The Yankee Group (Sep. 2003): In 2003, $84 million PTT revenue, 2.3 million

PTT subscribersBy 2008, $10.1 billion PTT revenue, 340 million

PTT subscribers

Future Cellular Trends

Richer groupware applicationsData+ voice+ video

Adopting TCP/IP infrastructure Session Initiation Protocol (SIP) signalingOMA, 3GPP, 3GPP2 standards

Wireless Networks Trends

Maturing standards (Wi-Fi, WiMAX) High availability of hot spots Supported in PDAs, Pocket PCs, laptops,

and cellular-phones Emerging standards and working groups:

IETF-MIP, Open Mobile Alliance, 4G

Mobile Networks Trend

Groupware popularity +

Wireless access

availability

B3G Convergence

Cellular going IP +

The Future Network IP based

Cellular N et .

T he I nternet

AccessPoint

AccessPoint

M obile U ser

M obile U ser

S tat ionary U ser

Cell

Mobile Groupware Design Goals

Mapping group names to subscribers Mobility support Seamless handoff QoS support for RT applications Transport efficiency Transport reliability Roaming, AAA Scalable Support for incremental deployment

Previous Solutions: Mobile IP RFC 3344

The standard for seamless mobility Unicast sessions only Dependence on a single home Triangle routing

Poor performance and lack of QoS support Inadequate for RT applications

Route optimization to 3344 Eliminate triangle routing Difficult to deploy Unclear if it can support simultaneous movements

Previous Solutions: Cellular

OMA PoC working group (cellular operators) Single server architecture Large overhead Triangle routing

Our Proposed Solution

Mobility and Group Management Mobility and Group Management ArchitectureArchitecture

[Lavi, Cidon, Keidar MWCN 2004][Lavi, Cidon, Keidar MWCN 2004]

MaGMA’s Architecture

Consists of Mobile-Group Managers (MGMs) and Mobile Nodes (MNs)

Version 1: MGMs static and well-known MGM in charge of one or more domains Entering a new domain, MNs obtain IP

addresses and contact local MGMs

Example: MaGMA Groups

M GM 5

M GM 4

M GM 2 M GM 3

M GM 1

M N 2

M N 3

M N 1M N 4

M N 5

M N 6

Domain-4

Group Blue Group

Red

Group Management Approaches Subscription model

Sending MN implements multicast MGMs provide list of subscribers in group MGMs notify sender of changes

join, leave, move (change IP) Good for lightweight servers, small groups

Multicast overlay model MGMs implement multicast + QoS + reliability using

transport-level overlay Scalable in group size, good for low battery clients

Group Management inSubscription Model

MaGMA Group View

T he I nternet

M N 1

M N 4

M N 2M N 3

M N1->D1M N2->D2M N3->D2M N4->D3

M N2->D2M N3->D2M N1->D1M N4->D3

M GM 1

M GM 2

M N4->D3M N1->D1M N2->D2M N3->D2

M GM 3

GROUP X

MaGMA Group ViewGROUP X

T he I nternet

M N 1

M N 4

M N 2M N 3

M N1->D1M N2->D2M N3->D2M N4->D3

M N2->D2M N3->D2M N1->D1M N4->D3

M GM 1

M GM 2

M N4->D3M N1->D1M N2->D2M N3->D3

M GM 3

MaGMA Solutions

MGMFlood: Flood all events (join, move,…) to all MGMs Sends unnecessary control messages to

MGMs not in group MGMLeader: forward group events only to

MGMs participating in the groupLess control overhead

MaGMA Solution 2:MGMLeader

M GM

M GM

M GM

M GM

M N

M GM M GM

M GM

join/move

/leave

group X

M N

join

Needs the group’s

view

Coordinator

view

View Consistency with Concurrent Joins

MGM1Coordinator

MGM2 MGM3MN1join

MN2 join

view

view

Ignore

join(MN1)join(MN1)

join(MN2)

join(MN2){MGM2, MGM3}

Solution uses a Local Event Counter (LEC) per MGM

[2,1,1]

[1,1,2]

local view

{MGM2, MGM3, MGM1}

Inconsistent

LEC1=1

LEC1=2

Handling Move Goal 1: smooth handoff Goal 2: reduce control overhead

while keeping view consistency

M GM 6 M GM 5

M GM 4

M GM 7

M N

M GM 2 M GM 3

M GM 1

group X

Coordinator

move from MGM1

move

view

transport tunnel

Coordinator Election

Need to ensure a single coordinator Need to address coordinator leave See MWCN paper…

Some Simulation & Analysis Results

Ns2 Simulations & Analysis: MGMFlood vs. MGMLeader Control Overhead Evaluation

Simulation: MaGMA vs. MIP Transport Delay

M N

M GM 0

M GM 2M GM 1 M GM 3

source

5Mb 20ms

MGM1 functions as the HA

MaGMA Multicast Overlay Model

MGMs organized in overlay Multicast data forwarded over the overlay

Ongoing and Future Direction

Efficient solutions for multicast overlay modelKeeping MGM-level views not MN-level viewsOptimizing overlay, adding QoS supportMapping groups to optimal servers

Fault-tolerance: tolerating MGM failures and dynamic changes

Advanced application support

Conclusions

Wireless networks (Wi-Fi, WiMAX) will merge with the Internet and cellular infrastructureConverged B3G will be IP-based

Users will demand support for real-time (RT) groupware such as PTT

Current mobility solutions - inadequate for RT MaGMA can provide comprehensive support

for mobility, group management, and QoS