1. No category

Towards Emergency Networks Security with
Per-Flow Queue Rate Management
Maurizio Casoni, Carlo A. Grazia, Martin Klapez, Natale Patriciello
Department of Engineering Enzo Ferrari
University of Modena and Reggio Emilia
TCenter
St.Louis, 27 March 2015
International Workshop on Pervasive Networks for
Emergency Management PerNEM’2015
C. A. Grazia (PhD Student)
Active Queue Management
27 March 2015
1 / 15
Introduction
Buffer management
Results
Conclusions
The PPDR-TC project: Public Protection and Disaster
Relief - Transformation Center
PPDR-TC goals
Effective Public Protection & Disaster Relief (PPDR) communications
Preparation of the next generation of PPDR systems
The Consortium:
C. A. Grazia (PhD Student)
Active Queue Management
27 March 2015
2 / 15
Introduction
Buffer management
Results
Conclusions
Talk overview
1
Introduction
Problem
Real Example
State-of-the-art
2
Buffer management
Description
3
Results
4
Conclusions
C. A. Grazia (PhD Student)
Active Queue Management
27 March 2015
3 / 15
Introduction
Buffer management
Results
Conclusions
Problem
what
to support PPDR communications
QoS guarantees
attack prevention
why
resources are precious after a disaster
satellite tech are often the only one solution (TCP problems)
malicious users make things more challenging
where
cooperative network layer: buffer management
C. A. Grazia (PhD Student)
Active Queue Management
27 March 2015
4 / 15
Introduction
Buffer management
Results
Conclusions
Problem: simple PPDR scenario
C. A. Grazia (PhD Student)
Active Queue Management
27 March 2015
5 / 15
Introduction
Buffer management
Results
Conclusions
Effect of an attack over cooperative environment
14
Throughput (Mb/s)
12
10
8
6
4
2
0
0
10
20 30 40
Time (s)
Attacker
Flow 1
C. A. Grazia (PhD Student)
50
60
0
10
Flow 2
Flow 3
Active Queue Management
20 30 40
Time (s)
50
60
Flow 4
27 March 2015
6 / 15
Introduction
Buffer management
Results
Conclusions
State of the Art
typical solution
Network layer techniques: AQM or Packet Scheduler
Scheduling is useless with same traffic type
In satellite environment, TCP congestion is the critical point
weaknesses
difficult to bound the bandwidth
packets in the queue
−→
IP level
bandwidth (congestion control)
−→
TCP level
how to move from packets burst to bandwidth?
flooding attack cannot be managed by packet schedulers
C. A. Grazia (PhD Student)
Active Queue Management
27 March 2015
7 / 15
Introduction
Buffer management
Results
Conclusions
AQM classical schema
C. A. Grazia (PhD Student)
Active Queue Management
27 March 2015
8 / 15
Introduction
Buffer management
Results
Conclusions
AQM classical schema
C. A. Grazia (PhD Student)
Active Queue Management
27 March 2015
8 / 15
Introduction
Buffer management
Results
Conclusions
Proposed solution: Queue Rate Management (QRM)
simple AQM placed at networking layer
born for cooperative networks (node-rate given)
RCP-AC, XPLIT, ECN, CCML (Satellite, Data Centers, etc)
malicious node could exceed it
QRM node-rate hypothesis not strong
it traces packets to get the flow RTT and calculate the
actual flow rate
C. A. Grazia (PhD Student)
Active Queue Management
27 March 2015
9 / 15
Introduction
Buffer management
Results
Conclusions
QRM benefits
1
gives control about length and bandwidth in a single
queue manager
2
deterministic drop policy
3
consume O(n) memory like standard AQM
4
5
time complexity of O(n), simulations show a Θ(1)
behavior (we already have an O(1) version)
max queue length is bounded at BDP value
(Theorem)
C. A. Grazia (PhD Student)
Active Queue Management
27 March 2015
10 / 15
Introduction
Buffer management
Results
Conclusions
PPDR case study
LTE BandWidth: 20Mbit/s
Satellite bandwidth
20Mbit/s
Satellite delay 350ms
Queue BDP size of 1.8MB
[2, 32] FRs involved:
[1, 16] Malicious users
C. A. Grazia (PhD Student)
Active Queue Management
27 March 2015
11 / 15
Introduction
Buffer management
Results
Conclusions
CoDel vs QRM: Rate and fairness
12
12
10
10
Throughput (Mb/s)
Throughput (Mb/s)
5 ns-3 client nodes, 1 node is an attacker
8
6
4
2
8
6
4
2
0
0
0
10
Attacker
Flow 1
20
30
Time (s)
Flow 2
Flow 3
C. A. Grazia (PhD Student)
40
50
60
0
Flow 4
Active Queue Management
10
Attacker
Flow 1
20
30
Time (s)
Flow 2
Flow 3
40
50
60
Flow 4
27 March 2015
12 / 15
Introduction
Buffer management
Results
Conclusions
1500
1500
1000
1000
500
500
0
0
10
20
30
Time (s)
40
50
0
60
2000
2000
1500
1500
1000
1000
500
500
0
0
Good Flow RTT
Max Queue Occupancy
C. A. Grazia (PhD Student)
10
20
30
Time (s)
40
50
Queue Occupancy (KB)
2000
RTT (ms)
2000
Queue Occupancy (KB)
RTT (ms)
No buffer management vs QRM: RTT and queue length
0
60
Good Flow RTT
Max Queue Occupancy
Active Queue Management
27 March 2015
13 / 15
Introduction
Buffer management
Results
Conclusions
QRM Scalability: queue length
Gateway Queue Occupancy (KB)
2000
1800
1600
1400
1200
1000
800
600
400
200
0
2
4
DropTail
CoDel
C. A. Grazia (PhD Student)
8
# Nodes
16
32
Red
QRM
Active Queue Management
27 March 2015
14 / 15
Introduction
Buffer management
Results
Conclusions
Conclusions
Buffer Manager: QRM
a novel timestamps based AQM for infer and bound the
flow rate
efficiently insulate malicious traffic (flooding attack)
effective use of the typical BDP standard buffer-size
optimal run-time time and space complexity
preserves all the cooperative feature (QoS, Latency, etc)
C. A. Grazia (PhD Student)
Active Queue Management
27 March 2015
15 / 15
thank you
for your attention
[email protected]
extra slides
QRM Defense Scalability: queue length with attackers
growth over 32 total nodes
Gateway Queue Occupancy (KB)
2000
1800
1600
1400
1200
1000
800
600
400
200
0
1
2
DropTail
CoDel
4
# Attackers
Red
QRM
8
16
Normalized Final Goodput Good Node
QRM Worst-case Tx Data: good nodes
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
2
4
DropTail
CoDel
8
# Nodes
16
Red
QRM
32
Normalized Final Goodput Good Node
QRM Worst-case Tx Data: good nodes with attackers
growth over 32 total nodes
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
1
2
DropTail
CoDel
4
# Attackers
Red
QRM
8
16