DDoS Defense by Offence

1
DDoS Defense by Offence

• Michael Walfish, Mithili Vutukuru, Hari
  Balakrishnan, David Karger, and Scott Shenker
• MIT CSAIL, UCB and ICSI
• ACM SigComm 2006

2
Server Capacity and Traffic

• Attackers traffic B requests/s
• Good users traffic G requests/s
• Servers normal load g requests/s
• Servers capacity C requests/s
• CG/(GB) gt g ? C gt g(1B/G)

3
How to Deal With Requests with Different Server
Load

• Time Slice each request
• Normal 1 slots
• Need database service extra 4 slots
• Need parallel search extra 2 slots
• etc
• For a req that needs K time slot
• The user has to win the auction K times to
  process this req

4
Server Capacity Effect
5
Bottleneck Link Effect

• Drawbacks of the protocol
• ????? ?????
• 10 good nodes and 10 bad nodes ? No bottleneck
• 30 nodes with bottleneck 40Mbps

6
Peer Heterogeneous Effect (1/2)

• Peer with shorter RTT tent to have large
  proportion of the server

7
Peer Heterogeneous Effect (2/2)

• Peer with larger b/w tent to have larger part of
  the server

8
Conclusion

• How do I think of this paper
• Easy to read, nice idea
• Easy to combine with other schemes
• Something insufficient
• The test scale is too small
• Does the assumption really holds
• The method may consume large b/w