4.6 SDN Traffic Engineering 105
(i.e., set ⟨source, sink, volume⟩s), and takes constraints, and objective
function as other inputs. Then, it determines multiple paths for
each flow (e.g., k-shortest paths) and computes locally optimized
bandwidth allocations to them on their paths. The TE results (paths
and bandwidth allocations) are enforced and implemented in the
region using EnfApp.
4.6.1.3 Scalability Benefit
Intuitively, SdnTE can scale global flow optimizations, having exponential worst-case complexities [100]. It can distribute flows among
regions, and then semiglobally run TE in a recursive-parallel manner
from the root region to the leaf regions. At a given level, cNodes can
perform their TE on small set of flows in a small search space in parallel. Recursively, this is followed by parallel TE in the level as follows.
In Figure 4.16, cN7’s TE is followed by that of cN5 and cN6 in parallel
and then parallelism among cN1–4.
4.6.2 Design Challenges
A close inspection reveals our design of SdnTE has a few challenges.
In the literature, there are hierarchical routing mechanisms (e.g.,
PNNI [101] and Nimrod [102]) in recursively aggregated networks.
However, these systems do not provide solutions to our design
challenges because SdnTE is a centralized, multipath TE solution in
the SDN environment. In contrast, they are distributed, single-path
routing protocols in nonprogrammable networks. These systems
do not optimize flows, mostly deal with the path computation, and
at best are equipped with simple circuit reservation mechanisms.
Compared to them, SdnTE is challenging as it needs to semiglobally
plan flows based on the ISP-selected objective functions such that
network resources are highly utilized. In particular, these routing
systems are inefficient from TE aspects. In general, poor network
utilization, circuit reservation failures, and congestion are inherent to
them. We now elaborate on the SdnTE challenges.
Challenge #1 – optimized TE in the presence of recursive abstractions: In the flat SDN, a single TE application has full control on
switches, flows, and traffic; it can compute globally optimized paths
and bandwidth allocations by running linear flow optimizations
(e.g., [99, 104]). For TE scalability, SdnTE exposes a partial and
(i.e., set ⟨source, sink, volume⟩s), and takes constraints, and objective
function as other inputs. Then, it determines multiple paths for
each flow (e.g., k-shortest paths) and computes locally optimized
bandwidth allocations to them on their paths. The TE results (paths
and bandwidth allocations) are enforced and implemented in the
region using EnfApp.
4.6.1.3 Scalability Benefit
Intuitively, SdnTE can scale global flow optimizations, having exponential worst-case complexities [100]. It can distribute flows among
regions, and then semiglobally run TE in a recursive-parallel manner
from the root region to the leaf regions. At a given level, cNodes can
perform their TE on small set of flows in a small search space in parallel. Recursively, this is followed by parallel TE in the level as follows.
In Figure 4.16, cN7’s TE is followed by that of cN5 and cN6 in parallel
and then parallelism among cN1–4.
4.6.2 Design Challenges
A close inspection reveals our design of SdnTE has a few challenges.
In the literature, there are hierarchical routing mechanisms (e.g.,
PNNI [101] and Nimrod [102]) in recursively aggregated networks.
However, these systems do not provide solutions to our design
challenges because SdnTE is a centralized, multipath TE solution in
the SDN environment. In contrast, they are distributed, single-path
routing protocols in nonprogrammable networks. These systems
do not optimize flows, mostly deal with the path computation, and
at best are equipped with simple circuit reservation mechanisms.
Compared to them, SdnTE is challenging as it needs to semiglobally
plan flows based on the ISP-selected objective functions such that
network resources are highly utilized. In particular, these routing
systems are inefficient from TE aspects. In general, poor network
utilization, circuit reservation failures, and congestion are inherent to
them. We now elaborate on the SdnTE challenges.
Challenge #1 – optimized TE in the presence of recursive abstractions: In the flat SDN, a single TE application has full control on
switches, flows, and traffic; it can compute globally optimized paths
and bandwidth allocations by running linear flow optimizations
(e.g., [99, 104]). For TE scalability, SdnTE exposes a partial and