The and uses it to split traffic. The

The flow-based load distribution models solve the problem by
assigning packets of the same flow to the same path. Although the risk of
packet reordering decreases, queuing packets over the same path causes the
end-to-end delay to increase. Flowlet Aware Routing Engine (FLARE) 14
proposed to group packets into small subsets of packets called flowlets, and to
use these flowlets as the scheduling module. Flare defines a threshold time
that assists to remove packet re-arrangement. One disadvantage of FLARE is that
it does not consider packet loss. Packet loss is important when transmitting
TCP traffic because FLARE may re-transmit packets in saturated path. Another
approach was presented in 15. In 15 the authors investigate the challenge
of splitting a traffic flow over the worldwide interoperability for microwave
access (WiMAX) and the local area wireless computer networking technology
(Wi-Fi) links and proposed an airtime-balance method. This method maps the
traffic load to an airtime cost function and uses it to split traffic. The idea
is to send packets to radio nodes so that the airtime cost is balanced.
However, this solution does not deal with packet re-arrangement. Adaptive
Load Balancing Algorithm (ALBAM) 16 focused on the TCP drawbacks found in
FLARE and suggested an algorithm that schedules traffic only when the packet
inter-arrival time is able to neutralize the path delay difference. Despite the
fact that accurate delay estimations must be available, ALBAN does not evaluate
the overall end-to-end delay.The packet-based load balancing models
manage to reduce the end-to-end delay, by using packets as the basic allocation
unit. This strategy reduces their ability to maintain low levels of packet
reordering. Effective Delay Controlled Load Distribution (E-DCLD) 19 realized
that inefficient load balancing can degrade the network performance and tackles
the problem by developing an optimization problem that balances the end-to-end
delay among all the available paths. One of the main concerns of E-DCLD is the
low convergence time. Sub-Packet Multipath Load Distribution (SPMLD) 22
proposed to formulate the problem as a constrained optimization problem that
minimizes the end-to-end delay. Its main idea is to reduce packet re-arrangement
by grouping multiple paths into a single virtual path. One of the main concerns
of SPMLD is its complexity, since it proposes two distributed algorithms that
have to be implemented in the source and the destination nodes.