Interaction between Control and Data Planes
The widespread use of the Internet as well as its potential for
disruptive effects on both business and society has made the Internet
one of the most important communication infrastructures.
All IP services (including domain name service or DNS, web hosting, and email)
depend on connectivity that is built on the routing infrastructure.
The Internet traffic has exhibit increasing variability in both
data plane and control plane. In data plane, malicious attacks
such as worm or virus scan can impact the dynamics of
the traffic, while in control plane, both malicious attacks
and unintentional misconfigurations can impact the stability and
reliability of the Internet. Router performance under
variable data and control traffic load is critical for understanding
the robustness and performance of the Internet infrastructure.
Variablity of data traffic:
How does the variablity of data traffic impact the performance of routers?
The variablity includes packet size, packet interarrival time, and
packet type. Systematic studies of potential variablity can further
facilitate the understanding of variablity of control plane as described
below.
Variablity of control plane traffic:
The variablity of control plane traffic (generated by both interdomain and
intradomain routing protocols) can be caused by attacks
on either data plane or control plane. For example, evidences have shown
that worm traffic has caused BGP session reset or router reboot, which
in turn leads to large variablity on control plane.
Further, unintentional human errors or intentional misconfigurations
can lead to persistent variablity on control plane.
Although large scale exploitation of routers has not been reported yet,
the potential impact of these attacks can be so large that preventive
measures must be taken in the near future.
Interaction between control and data plane traffic:
The instability of control plane can further impact the performance of
data plane. Updating routing and forwarding tables consumes a large
number of CPU cycles when there is a significant amount of routing update
traffic. This might lead to significant performance degradation of data
forwarding, in particular to low-end routers. Further, to what extent can
the instability of control plane impacts delay, loss on data plane?
Are loops caused by transit behavior of routing protocols or misconfiguration?