Document

Characteristics of Denial of
Service attacks on Internet
using AGURI
Ryo Kaizaki
Keio Univ. ,Japan
[email protected]
Goal : support of network
operation against DoS attacks
• There are many DoS(Denial of Service) attacks
(ex)slammer worm in 25 Jan.
• There are many types of attacks
→AGURI : design & implementation of the traffic profiler
• AGURI
– single & range target
– flexible detection
• Observation on WIDE(AS2500) backbone
• Report of DoS attacks and their characteristics
CNN
,25 Jan 2003
Focus : types of DoS attacks
type
victims
Application
Logic attacks
Operating System
DoS
attacks
Flooding attacks
Resources of an end node
- CPU , memory, network I/F
Resources of router
- CPU & I/F, bandwidwh
Flooding attacks
Attacker
Host A
Server
Router A
Router D
Host B
Router C
Host C
Router B
Flooding attacks
•Attacker sends massive packets
Attacker
Host A
Server
Router A
Router D
Host B
Router C
Host C
Router B
Flooding attacks
•Router C drops packets.
Attacker
Host A
Server
Router A
Router D
Host B
Router C
Host C
Router B
Drop packets
Network operation against
flooding attacks
1.Detection
Is network
in trouble?
Attacker
Host A
Server
Router A
Router D
Host B
Router C
Host C
Router B
Drop packets
Network operation against
flooding attacks
2. Detection of
victims
Attacker
Host A
Server
Router A
Router D
Host B
Router C
Host C
Router B
Drop packets
Network operation against
flooding attacks
3. Attacker’s packets
are the packets!
Attacker
Host A
Server
Router A
Router D
Host B
Router C
Host C
Router B
Drop packets
Network operation against
flooding attacks
4. Drop attacker’s packets
drops
packets
Attacker
Host A
Server
Router A
Router D
Host B
Router C
Host C
Router B
Drop packets
Filter expression
against flooding attacks
• Simple flooding attacks
deny ip hostA port 100 hostB port 200 tcp
→we can use single expressions.
• Flooding attacks to a company/campus/ISP
deny ip hostA port 100 10.0.0.0/24 port 200 tcp
→we can use range expressions.
→
best : drop only attacker’s packets.
better : drop some packets including attacker’s.
worst : do nothing
Type of attacks
(simple flooding attacks)
target
tuples
Source IP address
Destination IP address
Source port number
Destination port number
Protocol
single
range
random
Type of attacks
(port scan)
target
tuples
Source IP address
Destination IP address
Source port number
Destination port number
Protocol
single
range
random
Type of attacks
(attacks to network)
target
tuples
Source IP address
Destination IP address
Source port number
Destination port number
Protocol
single
range
random
Type of attacks
(source spoofing)
target
tuples
Source IP address
Destination IP address
Source port number
Destination port number
Protocol
single
range
random
Types of attacks
• There are many types of attacks
– no characteristics in source IP address
– no characteristics in destination port number
– characteristics of destination IP address in range
→ for monitoring attacks,
needs on various point of views
General methods
• Rule based matches
– Rule based matches with pre-defined rule sets
(ex) IDS
• Flow based aggregation (single)
(ex) Cflowd , Netboy
• AS based aggregation (range)
– Skitter(arts++)
AGURI’s concept
• Break 5-tuples to each element
– Enable to detect flooding attacks using
characteristics of a element.
• Aggregation each element
– Enable to detect flooding attacks
• Simple target
• Range target
Design of AGURI
10.0.0.0/29
• Put address
information on
binary tree
structure
10.0.0.0/30
10.0.0.0
.1
.2
.3
10.0.0.4/30
.4
.5
.6
.7
Design of AGURI
• Patricia tree
• LRU
• threshold
AGURI’s output
•profiles
•src_adr
•dst_adr
•src_port
•dst_port
[src address] 4992392382
0.0.0.0/0
87902964
60.0.0.0/6
97928228
62.52.0.0/16 51875058
64.0.0.0/8
100831910
64.0.0.0/9
74610984
128.0.0.0/2
142349668
133.0.0.0/8
69142535
150.65.136.91 54123094
:
:
:
(100.00%)
(1.76%/100.00%)
(1.96%/3.00%)
(1.04%/1.04%)
(2.02%/3.51%)
(1.49%/1.49%)
(2.85%/13.33%)
(1.38%/1.38%)
(1.08%)
Measurement on WIDE backbone
• Data A : 9months
• Data B : 3months
• Data C : 15months
Router A
Switch A
Data A
Router C
Data C
Switch B
Router B
ISP
Data B
ISP
US
JPN
Characteristic of attacks in time series
(destination address)
host 1
host 2
host 3
host 2
(result1)
Source spoofing attacks
(destination address)
host 1
(result 1)
Source spoofing attacks
(source IP address)
128.0.0.0/2
(result 1)
Source spoofing attacsk
target
tuples
Source IP address
single
range
random
Destination IP address
Source port number
Destination port number
Protocol
→ drop packet which destination ip address is victim
(result 2)
port scan
•IPv4
•TCP
•dst prot
•Begin port number
•++3
[ip:proto:dstport] 10933438650 (100.00%)
0/0:0:0 50394643 (0.46%/100.00%)
4:6:0/0 123970078 (1.13%/96.16%)
4:6:0/3 136730580 (1.25%/95.03%)
4:6:0/10 110321675 (1.01%/51.22%)
4:6:0/12 180612063 (1.65%/11.77%)
4:6:2 220337940 (2.02%)
4:6:5 220259760 (2.01%)
4:6:8 224630700 (2.05%)
4:6:11 220901820 (2.02%)
:
2
:
4:6:104 229349040 (2.10%)
4:6:107 220964460 (2.02%)
4:6:110 221768098 (2.03%)
4:6:119 213498789 (1.95%)
(result 2)
port scan attack
target
tuples
Source IP address
single
range
Destination IP address
Source port number
Destination port number
Protocol
→ drop packet port / destination in range
random
(result3)
Slammer worm
(source IP address)
128.0.0.0/3
(result 3)
Slammer worm
(destination IP address)
128.0.0.0/1
(result 3)
Slammer worm
(Destination port number)
4:17:1434
(result 3)
Slammer worm
target
tuples
Source IP address
single
Destination IP address
Source port number
Destination port number
Protocol
→ drop any any eq 1434 udp
range
random
conclusion
• Flooding attacks : use up network resources
• AGURI
– Can detect attacks from single target to range target
• Measurement on WIDE backbone
• Detect many types of flooding attacks
– Drop flooding attack’s packets at routers.