|
27.
01.
2006
PeerReview: Detecting Deviant Behavior in Distributed
Systems
Andreas Haeberlen
Max Planck Institute for Software Systems, Saarbrücken
Distributed systems have complex failure modes and are prone
to a variety of security attacks. In one class of attacks,
a node is compromised by an attacker and then deviates from
the protocol in order to exploit or disrupt the system, deny
services to users, or to censor content.
Byzantine state machine replication can mask faults of this
type as long as less than a third of the nodes are faulty at
any given time. However, while these algorithms are very
powerful, they tend to have significant overhead and do not
scale to large groups.
In this talk, I will present a novel, complementary technique
that cannot mask Byzantine faults, but can quickly expose them.
Once a fault is exposed, the corresponding node can be removed
from the system, and its functions can be reassigned to other
nodes, which limits the disruption that is caused by the fault.
|
