China Communications • February 2017
106
network.
Lecture [24] takes Deep Packet Inspection
(DPI) as a critical building block for the mid-
dleboxes. In its proposed orchestration struc-
ture, the trafc of any policy chain must rst
through a DPI before it is routed to other mid-
dleboxes. Our research is orthogonal with it,
since our optimized FCM model can be used
in the DPI-specialized policy chain’s orches-
tration.
Qazi et al. [25] provide a SIMPLE policy
enforcement layer based on SDN to build the
service chain. It focused on chaining middle-
boxes based on programmable switch. Our
model is optimizing the selection of instances
for each network function. Literature [26] is
similar with [25]. It dene an amazing owtag
with causal context. Our work could use this
tag as the one of the identiers for the network
function.
IETF started the group of service func-
tion chain (SFC) architecture to decouple the
physical topology and service function de-
ployments [27]. In this group, several drafts
have been submitted to show many user cases.
They are focused on packet structure, security,
control plane design etc. As stated in [28][29],
researchers should promote some drafts to op-
timize SFC conguration.
Our work is built on the prior works of [2]
and [13] but differing from them for two as-
pects. Firstly, we re-abstract the low-level net-
work functions into ner units, network func-
tions, and expose them as the basic compo-
nents. This can be in favor of standardization
on northbound interface of SDN controller.
Secondly, we extend the modular composition
in the global network, and theoretically model
the scenario as Boolean linear programming.
III. PRELIMINARIES
In this section, we will briey discuss the ba-
sic conceptions about network functions, com-
bined chain and policy before presenting the
FCM.
[15], bring up. A mass of schemes have been
provided and developed from multi-thread
controllers, like Maestro[16] and Beacon
[17], to the distributed system, like Hyper-
Flow[18], Onix [19], SiBF [20], and Devolved
Controllers [21] Kandoo [22] and ElastiCon
[23] provide logically centralized but physi-
cally distributed control planes. Most of those
works has taken more attentions on scalability
and reliability. Alternatively, our work in this
paper focuses on organizing various network
functions by abstracting basic network func-
tions.
Glen Gibb et al. [2] present an architecture
for add network function via outsourcing to
external feature providers who is the third-part
capable companies. This architecture has the
purpose of ourishing the network function in-
novations by introducing competitions among
providers. However, outsourcing will bring
more instances deployed by different provid-
ers that have similar or overlapped functions
inevitably co-exist in the network. Unfortu-
nately, this work does not point out the deci-
sion-making about the choice of appropriate
instances from candidate to form a completely
optimal processing pipeline.
Then, another work, called Beacon [8],
aims to develop an interface-friendly frame-
work and the runtime ability to start and stop
existing and new applications. Although Bea-
con indicates us to a more modularized ap-
proach for network functions, it is short about
how to operate them in the cooperative ways.
At the high level, Frenetic [11] and Pyretic
[12] give a composing language to manually
arrange the application in SDN, but they pres-
ent more stress on the rapid innovations. J. C.
Mogul [13] et al. provide a design for control
plane, called Corybantic, which supports com-
position of modular programs in independent
controller and achieves both modular compo-
sition and optimization, while it limits itself in
single controller and formulates the controller
stack at a high layers. Our system decouples
orchestral plan from instances that actually
execute the plan, and associates through our
FCM model to select instances distributed the