Resource poolability
Abstract.
There are moves in the Internet architecture community to add
multipath capabilities to TCP, so that end-systems will be able to
shift their traffic away from congested parts of the network.
What features of a network topology or routing protocol assist in this?
I will describe a metric, the poolability index, which measures for each resource
how easy it is for traffic to be shifted away from that resource e.g. in the event of a traffic
surge or link failure.
Key words.
multipath TCP, resource pooling, multipath routing, load balancing
A network with four shared resources and four traffic sources, three of
which have a choice of path. The resources have capacities
C1=1,
C2=2,
C3=3,
C4=5.
Each resource is colour-coded according
to its
resource poolability Ψ:
- If resource 1 experiences a traffic
surge then traffic from sources
s1 & s2 can easily be
rerouted over other paths, which increases load on link 4,
so some traffic from s3 will be rerouted as a knock-on effect.
Traffic from s4 has no other paths available, but
it doesn't suffer as much as it might, because
s1 moves some of its traffic
from resource 1 to resource 3 which has more capacity available.
Mathematically, resource 1 has resource poolability
Ψ1,1=0.91.
-
If resource 4 experiences a traffic surge then
traffic from sources
s1 & s2 will have to use
resource 1 instead. But resource 1 has little capacity (as indicated by its small size), so
congestion will increase significantly. Mathematically, resource 4 has
resource poolability Ψ4,4=0.54.