An Intent-based Network Virtualization Platform
An Intent-based Network Virtualization Platform
for SDN
Yoonseon Han∗ , Jian Li† , Doan Hoang‡ , Jae-Hyoung Yoo§ , and James Won-Ki Hong∗†
∗ Division
of IT Convergence Engineering, POSTECH, Korea.
[email protected]
† Department of Computer Science and Engineering, POSTECH, Korea.
{gunine, jwkhong}@postech.ac.kr
‡ School of Computing and Communications, University of Technology Sydney (UTS), Australia.
[email protected]
§ Ministry of Science, ICT and Future Planning, Korea.
Abstract—Currently, the Software Defined Networking (SDN) management/operation. Currently, most of the SDN controllers
paradigm has attracted significant interests from industry and support OpenFlow as an essential South Bound Interface
academia as a future network architecture. SDN brings many (SBI) according to the Open Networking Foundation (ONF)’s
benefits to network operations and management including pro- specification [1].
grammability, agility, elasticity, and flexibility. With SDN and
OpenFlow, one of the promising SDN protocols, software defined Network Virtualization (NV) is a networking technology
Network Virtualization (NV) techniques can be designed and that creates dedicated Virtual Networks (VNs) over a physical
implemented via flow table segmentation to provision independent infrastructure. With the help of NV, multiple tenants are able
virtual networks (VNs). In this paper, we propose an intent based to share the underlying physical network resources, and they
virtual network management platform based on software defined can operate their isolated virtual networks independently. By
NV. The objective of the proposed NV platform is to automate the
provisioning VNs over the physical network, network function-
management and configuration of virtual networks based on high
level tenant requirement specifications, called intents. The design ality is abstracted from its physical elements. NV technology
and implementation of the platform is based on ONOS, an open- has the potential to reduce significantly CAPEX and OPEX
source SDN controller, and OpenVirteX, a network hypervisor. for network and network service providers with its flexible,
The platform is designed to provide multiple VNs over the same on-demand, and scalable provisioning capability. A possible
physical infrastructure to multiple tenants. approach to NV is the slice-based NV whereby a slice of
Keywords—Software Defined Networking (SDN), Network Vir-
the network physical resources can be allocated to a VN by
tualization, Intent Framework, Virtual Network Management and segmenting OpenFlows flow tables into partitions.
Configuration, Virtual Network Embedding A major deficiency of the slice-based NV approaches is
that they require underlying network infrastructure to be con-
I. I NTRODUCTION structed using OpenFlow protocol. Futhermore, the configura-
tion and management process of each VN are still complicated
Software-Defined Networking (SDN) is a new networking
and time consuming because of the lack of generally available
paradigm which enables flexible and efficient network manage-
VN embedding methods and automated VN provisioning pro-
ment. The essential principle of SDN is to decouple network
cesses. Currently, to configure and manage VNs, administrators
control and forwarding functions, and leave each function in
have to deal with all technical aspects of networking such as
its individual network plane. In the context of SDN, all control
underlying protocols, addresses, topologies, control rules, and
related functions are moved to a centralized control plane,
etc. To overcome this complexity and problems, in this work,
hence optimal network control decisions can be made using a
we introduce the principles of intent-based management. The
global networking view. SDN also provides the ability to sim-
definition of intent is not standardized yet, but it is generally
plify network design and operations, with this ability, we can
perceived as business or system level policies (or higher level
deploy complex network policies. To summarize, SDN brings
specifications). Intents is independent from specific network
four major features such as programmability, agility, flexibility,
technologies and vendor specific features. Moreover, it allows
and vendor neutrality to network management domain. By
the administrator to use higher level abstraction by using
properly utilizing those features, SDN has been promised to
business or system level terminologies and concepts. With
reduce CAPEX by using cheap, open, commodity switches
intents, users only need to concentrate on specifying what they
and cloud computing for replacing expensive middle boxes
need, rather than how to realize or implement the need.
and to reduce OPEX by providing simplified and centralized
In this paper, we propose an intent based VN management
This work was supported by the ICT R&D program of MSIP/IITP, platform for SDN to overcome these problems. The funda-
Republic of Korea. [B0190-15-2011, Korea-US Collaborative Research on
SDN/NFV Security/Network Management and Testbed Build], and [B0190- mental idea is to automatically manage VNs from high-level
15-2012, Global SDN/NFV OpenSource Software Core Module/Function tenant requirements using intents. To be more specific, the
Development]. proposed intent-based NV platform addresses the following
978-3-901882-85-2
c 2016 IFIP 353 ManSDN/NFV Full Paper
challenges: 1) using intent to express high-level VN require- directly programmed for the virtual elements to provide QoS.
ment specifications, 2) combining SDN and NV technologies Tenants can use their own specific controllers to control their
into a single framework, 3) automating the task of VN structure own VNs. Currently, available solutions include FlowVisor
composition and embedding. The platform will host multiple, [10], OpenVirteX [2], and FlowN [11]. However, the critical
independent, and isolated VNs, and support multi-tenancy. disadvantage of the software defined NV is that the physical
VNs belonging to different tenants may have different network network has to support SDN and OpenFlow.
configurations in terms of network address space, topology,
and may be governed by different policies. The proposed Another important related technology is Intent-based
platform is implemented on OpenVirtex network hypervisor mangement. Oxford dictionary defines “intent” as “an aim
[2] and ONOS SDN controller [3]. or plan or purpose.” The definition of intent for network
management is commonly perceived as business or system
level policies specified with common concepts and terminolo-
II. R ELATED WORK gies agreed by all related stake-holders. However, a clear
In this section, we focus on describing the overlay NV and concise definition of intent for network management has
and the software defined NV approaches that can be real- not been standardized yet. Several projects are proposed to
ized with SDN technologies. Most commercial NV solutions introduce intent for SDN application development and network
are based on overlay NV approach by leveraging tunneling management based on high-level requirements or management
or encapsulation techniques such as VMware NSX [4] and policies. Recently, intent-based interface has been pursued rig-
Microsoft Hyper-V [5]. Overlay NV approach can be further orously by IETF and major open-source project communities
categorized depending on whether the approach supports layer (ONF, ONOS and OpenDaylight) to provide a standardized
2 and layer 3 virtualization. Usually, to deliver packets, an intent-based northbound interface for SDN [3], [12]–[14]. The
ingress network device (switch or router) encapsulates packets specification methods was also developed by using language
by inserting an outer packet header indicating a specific virtual (e.g., NEMO [12], [15]) or policy graph (e.g., PGA [16]).
network instance ID (VVID). The encapsulated packets are
delivered to the destination according to forwarding rules, III. OVERALL P LATFORM A RCHITECTURE
and then, decapsulated to restore the original packets at the
egress network device. VXLAN [6] and NVGRE [7] are The proposed intent-based VN management platform is
two most representative overlay NV approaches that support designed to have a hierarchical architecture.The platform plays
layer 2 virtualization, while Generic Routing Encapsulation two roles which are 1) network hypervisor and 2) SDN
(GRE) [8] and Locator/Identifier Separation Protocol (LISP) controller. As a network hypervisor, the platform possesses VN
[9] are two most representative approaches that support layer management capabilities such as VN provisioning, modifica-
3 virtualization. tion, and removal, at the same time, it also provides interfaces
to relaying VN events and messages to external entities running
The advantages of the overlay NV approach include tenant specific applications. As an SDN controller, the platform
1) only network edges are involved in tunnel encapsula- provides network abstractions and control capabilities for both
tion/decapsulation, the remainder of the network remains physical and virtual networks. The overall architecture is
unchanged, 2) theoretically, unlimited number of VNs are depicted on Fig. 1. The design objectives are 1) to support
supported, 3) VNs are independent from the physical network multi-tenancy, 2) to provide network abstractions for applica-
topology and configuration, 4) VNs mobility can easily be sup- tion development, 3) to allow high-level tenant requirements
ported. As the overlay NV approach is based on encapsulation specification using intents, and 4) to support tenant specific
mechanism and tunneling, it also brings disadvantages. The controllers and applications by providing various interfaces.
main disadvantages include 1) Two separated networks, VN The platform has five layers: protocol adaptation, abstraction,
and PN, are maintained in terms of network services such as virtualization, virtual abstraction, and intent layer.
management, provisioning, and control, 2) it does not provide
mechanisms to provide to guarantee QoS, 3) it introduce high The protocol adaptation layer is responsible for processing
encapsulation and tunneling overheads, and 4) it incurs high protocol specific messages which are used to communicate
management complexity for both VN and PN at the same with physical hardware such as OpenFlow switches. The
time. To overcome these disadvantages, cloud platform such main roles of this layer are: 1) decoding protocol specific
as OpenStack provides several methods (such as Neutron) to messages and delivering them to proper providers located in
manage network resources. the abstraction layer, 2) managing communication channel be-
tween the controller and network devices, and 3) receiving the
With the introduction of SDN and OpenFlow technologies, requests from upper layers, encoding the request into protocol
it is possible to implement NV as an application or a service specific messages and transmitting to hardware devices. The
provided by an SDN controller via flow table segmentation. abstraction layer is responsible for abstracting protocol specific
This slice-based NV approach can support layer 1 - 4 net- concepts and hiding the details of underlying infrastructure.
work virtualization by matching appropriate packet headers. The abstractions can represent various elements and properties
Performance degradation caused by tunneling overheads is in a protocol-agnostic manner. Some representative abstrac-
eliminated. By inserting appropriate forwarding (flow) rules, tions are Device, Link, Topology, Event, Path, Flow, etc.
software defined NV approach can provide specific NV fea-
tures. Moreover, this approach introduces network abstractions The main responsibility of the virtualization layer is to
that can be utilized by management application such as virtual translate VN objects into physical objects by maintaining map-
links, virtual switches, and virtual routers. Within software ping information. The mapping information includes address
defined NV approach, corresponding physical hardware can be mapping and topology mapping. To support various strategies
Fig. 2. Three basic types of intents provided by the proposed platform as conflicted intents (possibly through duplication). The intents
that encounter any conflict are required to be modified and
negotiated into composable intents.
between endpoints. Various relations between endpoints such
as 1 to 1, 1 to many, and many to many, are possible.
D. Intent Mapping and Installation
- VN Chain Intent: This intent type is an extended
Our VN model representation consists of a set of net-
intent type from the VN Endpoint Intent to chain intermediate
work model objects and network behavior abstractions. To
network behaviors. This type of intent expresses network
embed a VN into existing physical network resources, virtual
service chains by composing virtual network functions or
model objects should be bound to actual physical objects.
physical middle-boxes.
The network behaviors also require to be translated from
To specify intents accurately, we need to define the context virtual network behaviors into installable physical network
that describes what, when, and how the specified intents should behaviors. The objective of VN embedding algorithm is to find
be applied. To express a context, an intent object requires an optimal mapping between the VN and physical resources
four attributes; resources, conditions, priority, and instructions. with considering to satisfy a set of requirements defined by
Resources refers to a set of virtual network objects involved network administrator. To efficiently embed VNs according
in intent specification. Conditions are a set of criteria that to management objectives, the platform may adopt various
describes when the intent will be activated. Priority is used to algorithms introduced in [18].
determine the execution order of intents. Instructions refers to
To translate VN model objects into physical objects, the
a set of actions that to be applied to the packets which satisfy
platform should bind all virtual entities into concrete network
resources and conditions what have been defined.
nodes. This process can be supported by managing tenant’s
end-points. Discovering and managing end-points of VNs are
C. Intent Composition and Conflict Checking challenging because the terminologies are not standardized
After specifying intents for each tenant requirement, the among stake-holders yet. To provide a solution, in this paper,
next step is to aggregate all intents to construct the requested we introduce the concept of “vocabulary store” which refers
VN model which consists of a set of network objects and to a knowledge store that contains information from various
behavior abstractions. The intent composition process is re- sources include 1) human domain experts and/or network
quired to translate high level specifications into a network administrators, 2) host and network discovery and 3) man-
driven concepts and terminologies. First of all, we need to agement protocols. To efficiently querying the equivalence
identify and manage VN endpoints. This is needed to unify and between entities, a ontology based knowledge representation is
translate concepts, resources, and terminologies into a single adopted for the platform to support a semantic based inference
unified form agreed by all involved entities. To address this mechanism. This does not mean that “vocabulary store” simply
issue, [17] proposed an end-point discovery protocol, while generates new knowledge by using inference rules, but it
[16] used “label” to apply policies. Note that “label” contains supports checking and finding simple equivalent relationships
a group of pre-defined endpoints with the same set of policies. between entities.
As outputs of intent composition, VN topology, address space
and policies that are used for governing the VN are computed.
The overall intent composition process is described in Fig. 3.
During the composition process, the proposed platform can
verify incorrect intents and detect conflicts between intents.
By inspecting intent resources, conditions, priorities, and in-
structions, an intent composer can detect conflicts. First of
all, we need to identify the relationship between intents to
check whether they have any dependency on each other. If Fig. 4. The role of vocabulary store
any of two intents are interdependent, the platform will lookup
the priorities of intent to check whether the instructions are
E. Intent Lifecycle Management
mutually inclusive. For instance, if two intents share the same
pair of source and destination addresses, and the identical To manage intents specified from multiple tenants, we have
instructions are defined in each intent, then they are detected designed a Finite State Machine (FSM) that represents the state