Content uploaded by Michail Alexandros Kourtis
Author content
All content in this area was uploaded by Michail Alexandros Kourtis on Jul 27, 2017
Content may be subject to copyright.
Performant Deployment of a Virtualised Network
Functions in a Data Center Environment using
Resource Aware Scheduling
Michael J. McGrath, Vincenzo Riccobene, Guiseppe
Petralia
Intel Labs Europe,
Leixlip, Co. Kildare, Ireland
Georgios Xilouris, Michail-Alexandros Kourtis
Institute of Informatics and Telecommunications,
NCSR “Demokritos”,
Athens, Greece
Abstract—The EU funded FP7 project T-NOVA, with the specific
goal of accelerating the evolution of NFV, proposes an open
architecture to provide Virtual Network Functions as a Service
(VNFaaS), together with a dynamic, and flexible platform for the
management of Network Services (NSs) composed by those
Virtual Network Functions (VNFs). The proposed architecture
allows operators to deploy distinct virtualized network functions,
not only for their internal operational needs, but also to offer
them to their customers, as value-added services. Virtual network
appliances (e.g. gateways, proxies or even traffic analyzers) can
be provided on-demand, eliminating the need to acquire, install,
and maintain specialized hardware at customer premises. This
demo illustrates early work carried out on the deployment of a
VNF on a Network Function Virtualization Infrastructure
(NFVI) using resource aware scheduling methods to ensure
optimal use of resources and performance.
Keywords—NFV, VNF, virtualisation, NFaaS, T-NOVA;
I. INTRODUCTION
Network Functions Virtualization (NFV) has received
significant interest an as approach that can address many of the
key challenges being experienced by service providers. These
challenges are being driven by an exponential growth in data
volumes with a corresponding fall in revenues per megabit.
Many service providers are currently investigating and in some
cases deploying VNFs to replace traditional high cost fixed
appliance based architectures particularly for edge of network
applications.
Virtualization is the key enabling technology that allows
traditional physical network functions to be decoupled from
fixed appliances by leveraging standard IT virtualization
technologies to consolidate various network equipment types
onto industry standard high volume servers, switches and
storage located in DCs. This approach will allow service
providers to innovate through the rapid deployment of new
services, increased customization and flexibility to meet
diverse customer needs, increased utilization of capital
resources, etc.
The EU funded FP7 T-NOVA project [1-2] is focused on
realizing the concept of Network Functions as a Service
(NFaaS) by designing and implementing an integrated
management architecture, for the automated provision,
management, monitoring and optimization of VNFs over
Network/IT infrastructures. As a technology, NFV
encompasses a wide variety of network functions, which have a
diversity of resource requirements. As a result T-NOVA is
carrying out research to develop an understanding of workload
types and their affinities for certain platform features and
technologies.
II. MOTIVATION
While virtualisation brings many benefits to Enterprise IT
and more recently to the Telecom domain it also brings many
challenges particularly in achieving the same level of
performance in comparison to the traditional fixed appliance
approach. The composition, configuration and optimization of
the virtualised resources are critical in achieving the required
levels of performance. Additionally given the origins of many
virtualisation technologies such as cloud computing
environments there are capability gaps that need to be
addressed in order to adequately support VNF/NS type
workloads.
Currently, within cloud environments resources are highly
abstracted, which again causes issues for the performant
deployment of VNFs and Network Services (NSs). For
example, it is important to expose the specific platform
features, such as unique CPU instructions and attached devices,
such as acceleration cards, co-processors or Network Interface
Cards (NICs) with advanced capabilities. Additionally many
hardware devices, such as NICs, have additional dependencies,
such as the availability of supporting software libraries - e.g.
DPDK - in order for a VNF to function in an optimal manner.
The demo presents initial work by the T-NOVA project to
address some of the limitations of cloud compute environments
with respect to scheduling of virtualised resources to host
VNFs in a performance manner.
III. DEMO DESCRIPTION
This demo illustrates how an NS with a single VNF (in this
case a network traffic classification VNF) can exploit the
platform requirements specification contained in an ETSI ISG
NFV VNF descriptor (VNFD) accompanying it, in order to
deploy and appropriately configure the virtualized
infrastructure resources in a performant manner as proposed by
the project. The Orchestrator parses the VNFD into
corresponding metadata are, which is then used to dynamically
construct a Heat template. The template is then used in an
OpenStack Cloud environment to automatically instantiate the
required virtualized resources with specific platforms features
that cannot be predictably provisioned using OpenStack’s
current scheduling mechanism. Finally a side-to-side
comparison of the performance of the virtual Traffic
Classification VNF’s running on ‘standard Virtual Machines’
(VMs) versus VMs with features to improve packet processing
performance is demonstrated using real-time instrumentation.
The network virtualization process starts at the lower levels
of the ISO/OSI protocol stack: VMs require vNIC/NICs to
provide connections to switches that in turn manage various
Virtual Local Area Networks (VLANs). Virtualization of
network functions typically involves two approaches; software-
assisted and hardware-assisted.
In software-assisted network virtualization,
communications are provided by hypervisors through virtual
switches (vSwitches) and protocols designed to coordinate and
manage virtualized network architectures at the edge of the
network. However, despite the flexibility of a software-assisted
approach, a significant disadvantage is the potential for the
virtual switch to act as a bottleneck with increasing number of
guests running on the host and network traffic volumes.
In hardware-assisted network virtualization, physical
hardware is directly assigned to the virtual guests in order to
both increase the performance and avoid the bottlenecks.
Intel’s suite of input/output virtualizations technologies, called
virtualization Technology for Connectivity (VT-c) is an
example of such an approach, which is complementary to
Intel’s VT-d (Virtualization Technology for Directed I/O). It
includes: Data Plane Development Kit (DPDK), Virtual
machine Device Queues (VMDq) and Virtual Machine Direct
Connect (VMDc). The latter is implemented using the PCI-SIG
standard called Single Root I/O Virtualizations (SR-IOV),
which enables a single PCI Express (PCIe) network adapter to
appear as many special-purpose adapters, called Virtual
Functions (VFs) that are available for direct presentation to
VMs, through VT-d.
The demo architecture shown in Figure 1 comprises of an
OpenStack Juno based cloud environment. The cloud
environment comprises of a controller and two compute nodes,
and a traffic generator connected on the same network domain
through a 10 Gbps switch. The compute node configuration
enables the VMs to use both vNICs connected to Open virtual
Switch (OvS) in the form of software-assisted solution and
physical NICs with PCIe passthrough/SR-IOV functionality in
the form of a hardware-assisted solution.
On the same host as the controller, a simulated Orchestrator
is running which receives a deployment request (which
includes an ETSI compliant VNFD) and converts it into
metadata representing the platform deployment requirements of
the VNF, which is used to dynamically generate a Heat
template. The template orchestrates the setup SR-IOV ports,
deployment of Traffic Classification VNF Components
(VNFCs) and configuration of the VNFCs The VNF used in
this demo comprises of two VNFCs, namely the DPI engine
and the Classification and Forwarding functionality. The
VNFCs are implemented and contained in the form of two
VMs. For the purpose of the demo, two different versions of
the VNFD are utilized: the first version contains no platform
specific features allowing the Orchestrator request a ‘standard’
deployment (based on OvS) of the DPI; the second version
contains specific platform features i.e. an SR-IOV capable NIC
which must be available on the physical sever were the VMs
are hosting the vDPI.
Once the two different instances of the Traffic
Classification VNF are running, traffic is generated by the
packet generator and sent to both the instances. The
performance of both deployments is compared using real-time
display of parametric data from the VMs captured using
embedded instrumentation agents in the VMs to highlight
differences in packet processing performance.
Fig. 1. Architecture of the demo testbed.
ACKNOWLEDGMENT
This work was undertaken under the Information
Communication Technologies, EU FP7 T-NOVA project,
which is partially funded by the European Commission under
the grant 619520.
REFERENCES
[1] T-NOVA Consortium, Deliverable 2.21- Overall system architecture and
Interfaces, June 2014, on-line:http://www.t-nova.eu/results
[2] G. Xilouris et. al, T-NOVA: A Marketplace for Virtualized Network
Functions, European Conference on Networks and Communications,
Bologna, Italy, June 2014.