Th`ese de Doctorat de l'université Paris VI Pierre et Marie Curie Mlle ...

Table 5.1: Pseudo-code specification of the Simple Dynamic Bandwidth Allocation algorithm (SDBA)
Algorithm SDBA(Kg, Rl, a l k )
(1) initiate all f n (2)
k =0, ∀ k ∈ Kg
remove from the link set L all links l ∈ L that have
a number of connections crossing them nl equal to 0
(3) for every link l ∈ L, calculateFl = Rl/nl
(4) identify the link α that minimizes Fα
i.e. α | Fα = mink(Fk)
(5) set f n k = Fα, ∀ k ∈ Kα, whereKα⊆Kgis the set
of greedy connections that cross link α
(6) for every link l, update the residual capacity and the
number of crossing greedy connections as follows:
Rl = Rl −
nl = nl −
k∈Kα f n k · al k
k∈Kα al k
(7) remove from set L link α and those that have nl =0
(8) if L is empty, then stop; else go to Step (3)
at the beginning of the first iteration of IDBA. For every successive iteration, a greedy
connection can be either marked as Served, when its total assigned rate (srk + f n k )isat
least equal to its offered load, or Non-served otherwise. Let K NS
g and K S g
of greedy Non-served and Served connections respectively.
denote the set
IDBA iterates over the two following steps until either all sources become Served or the
extra-bandwidth on the bottleneck links is completely allocated:
• First, the extra-bandwidth individuated in the network is allocated fairly among the
Non-served greedy connections using SDBA(K NS
g , Rl, a l k ).
• Second, for every Non-served greedy connection, k ∈ K NS
g if srk + f n k >Oln−1
k
r n k
is set equal to Oln−1
k
then
and the connection is classified as Served; the difference,
srk + f n k − Oln−1
k , is redistributed in the next iteration to the remaining Non-served
greedy connections.
46