Information Technology Reference
In-Depth Information
4 Simulation Results
We build a simulator to simulate the routing dynamics under our model. The
topology of the Internet and the AS relationships are inferred from CAIDA's
data set [7]. Although the inferred topology doesn't completely agree with the
actual Internet, this data set is the most complete and accurate one that is used
by present research. The number of IPs and prefixes in every AS are calculated
from the BGP routing tables collected by Route Views [8] and RIPE RIS [9].
Many ASes monitored by these two projects distribute in the core of the Internet,
so their routing tables cover almost all the routed IP prefixes in the world. More
precisely, we construct a connected network with 41204 ASes and 121310 AS
links. The tolerance parameter
α
is set to be 0.1, 0.3, 0.5, 0.7 and 0.9, representing
five levels of capacity of links. However, in fact, the capacities of links are various.
We simplify the situation in our simulator at first. Then we plan to differentiate
each link according to its position in the routing hierarchy in our future work.
The distributions of transmitted load of transit ASes and load of AS links
are shown in Fig. 3 and Fig. 4. As we can see, the Internet controlled by inter-
domain routing system exhibits a highly heterogeneous distribution of load. We
choose 79 candidates from the top 1% of transit ASes to construct core AS set,
and examine the rerouting dynamics in 20 time steps, denoted as
T
.Wecut
the AS link with highest load, denoted as
e
A
, to simulate the intentional attack,
whereas cut the link
e
B
at random to simulate the random breakdown. Then we
run policy-based BGP among ASes and measure its survivability under different
conditions. Next are some results.
10
20
10
15
10
10
10
5
10
0
10
1
10
2
10
3
10
4
Transit AS rank
Fig. 3.
Load of transit ASes
Fig. 5 and Fig. 6 are reachability on links with different capacities under
different initial failures. The first time step of every curve is the reachability
under the initial failure. Then it is followed by 20 time steps of consequent
reroutings. From the results we can see that no matter what kind of initial
failure is, the cascading effect on reachability of inter-domain routing system
is amplified when the tolerance parameter
α
is equivalent to 0.1. It's rational
to infer that the effect is also amplified when
α
is less than 0.1. Meanwhile it
is constrained into a very limited scope when
α
is equal to and greater than
