Distance-Dependent RED Distance-Dependent RED Policy (DDRED) Policy (DDRED) Sébastien LINCK, Eugen Dedu and François Spies [email protected] LIFC Montbéliard - France ICN07 – Martinique
Jan 16, 2016
Distance-Dependent RED Distance-Dependent RED
Policy (DDRED)Policy (DDRED)
Sébastien LINCK, Eugen Dedu and François Spies
LIFC Montbéliard - France
ICN07 – Martinique
ICN07 - Martinique, France 2
Overview
Introduction
Related Works
Proposition
Performance Evaluation
Conclusions – Future Work
ICN07 - Martinique, France 3
Introduction 1/2
Computer sends data to the highest throughput
TCP test the bandwidth by increasing its congestion
window (Cwnd)
A network is bandwidth-limited --> bottleneck
On the router which buffer the packets
A congested router loss packet
Router queue management can minimise the
congestion but cannot avoid it completely
ICN07 - Martinique, France 4
Introduction 2/2
TCP Losses
Loss --> packet retransmission
Retransmission --> network resources utilization
All the packets don't consume the same
resources
Idea : To favour some packets at time of
rejection
ICN07 - Martinique, France 5
Related Works
FRED [Lin & Morris 97]
Classification of flows according to their capacity to
answer the congestion
RTT and Cwnd size save on the router
AECN [Zheng & Kinicki 02]
RTT field in TCP header
On the routers : intervals of RTT and the
corresponding sub-queues
ICN07 - Martinique, France 6
Proposition 1/2
Proposition:
To favour a flow having traversed most of its way
How?
To reject, on the router queue, the packet nearest
from its source
Effect:
Better availability of the network resources
ICN07 - Martinique, France 7
Proposition 2/2
Policy based on RED mechanism
Don't change the principle but only the target
Distance parameter
New TCP option: initial TTL = TTLi
If ejection or ECN marking needed:
Find the nearest packet in the RED queue
Drop this packet instead of the incoming one
ICN07 - Martinique, France 8
Performance Evaluation 1/4
Simulation NS 2.30
Flower Network
A computer per router
500 TCP flows
Each flow
Size between 100KB
and 6MB
Source and destination
Start time
2 simulations/test: all
routers RED or DDRED
ICN07 - Martinique, France 9
Performance Evaluation 2/4
We always have losses but
Fewer than RED
Nearer to the source
ICN07 - Martinique, France 10
Performance Evaluation 3/4
Losses weighted by
the number of
consumed routers
Saved slots in RED
queue
Network resources
free for new flows
ICN07 - Martinique, France 11
Performance Evaluation 4/4
On average we save time
More than half of the flows are faster
ICN07 - Martinique, France 12
Conclusions – Future Work
New algorithm of rejection of packages taking
into account the traversed path
Time saved
More free network resources
Future work:
More realistic simulations with better traffic generator