Policy-based Congestion Management for an SMS Gateway

Alberto Gonzalez (KTH)

Roberto Cosenza (Infoflex)

Rolf Stadler (KTH)

June 8, 2004, Policy Workshop

Outline

• Intro to SMS

• Congestion Management in SMSG

• Policy-based Approach

• System Behavior

• Future Work

The Short Message Service (SMS)

• SMS is based on out-of-band message delivery. It permits users to send and receive text messages to/from their mobile phones

• Relevance– By the end of 2002, 30 billion messages exchanged

monthly– Growing at 0.5 billions per month. – SMS represents 10% of the revenue of mobile operators

• We consider two classes of SMS services– Guaranteed service (zero losses)– Non-guarateed service

• Bulk messages, information services

SMS Architecture

SMS Gateway Model

Routing

Engine

Inco

min

g P

orts

Out

goin

g P

orts

Acceptance Rate ThrottleMessage Drop

Congestion in a SMS Gateway

• Congestion can be caused by a persistent performance degradation of an outgoing port

• Why do we need congestion management ?– long buffers suppose a risk. The longer the queue, the

higher the cost of a system crash

• The focus of this work is to provide the EMG with congestion management capabilities that permit us to adapt dynamically to congestion

Related Workin Congestion Management

• Congestion management in routing engines has been extensively studied in the context of IP routers

• Our work differs from that field in – Problem space: congestion management for IP routers

considers physical networks. In contrast, an SMSG is a node in an overlay network, where the service rate of outgoing ports can vary, depending on the state of neighboring SMSGs

– Solution space: approaches to congestion management in IP networks often focus on flows. In contrast, flow-based mechanisms are not relevant in the SMS context, since an SMS message fits into a single packet

Addressing congestion in the EMG

• Two low-level mechanisms for reducing the load on a congested port – reducing the acceptance rate

• this reduction affects the loads on all outgoing ports. Therefore the overall throughput of the EMG is compromised

– dropping non-guaranteed messages that are routed to its associated queue

• They present a trade-off – Throughput vs losses

• The EMG manager has to choose between giving priority to:

• having low losses • having high throughput.

Routing

Engine

Inco

min

g P

orts

Out

goin

g P

orts

Acceptance Rate ThrottleMessage Drop

Functional Architecture for Realizing Congestion Control Policies

RoutingTableTraffic

Estimator

TrafficEstimator

TrafficEstimator

DropperOutgoing

queue

DropperOutgoing

queue

DropperOutgoing

queue

PDP

RoutingEngine

βjαi 3

σTijμj

μj

Oj

Low Losses High Throughputβmax

3

3

Iiαi

Outj

System behavior: unbalanced load

Utilisation

0

0,10,2

0,30,4

0,5

0,60,7

0,80,9

1

0 10 20 30 40 50 60 70 80 90 100

1 2 3 4 5

• Incoming traffic 6 m/s• Service rate 6 m/s• Non-guaranteed service: 90%

m/s

System behavior: balanced loadUtilisation

0

0,1

0,2

0,3

0,40,5

0,6

0,7

0,8

0,9

1

0 10 20 30 40 50 60 70 80 90 100

1 2 3 4 5

• Incoming traffic 6 m/s• Service rate 6 m/s• Non-guaranteed service: 90%

m/s

Summary

• Our architecture permits the EMG manager to control the tradeoff between high throughput of the EMG and low loss rate of the non-guaranteed service class

• In situations of congestion and changes in load pattern, the system dynamically re-configures, following the manager’s selected policy

• Our proposal does not need other SMSGs/SMSCs to include any type of congestion management

Future Work

• Service differentiation based on message sender, message content, etc

• Different algorithms for parameter estimation• Prototype evaluation using real traces