Professional Documents
Culture Documents
1 “Global cellular M2M technology forecasts and assumptions” March 2015, GSMA
2 “Ericsson Mobility Report” June 2015
www.spirent.com
1. Node-based virtualization: Orchestration
In this model, each one of the EPC
MME SGW PGW
components is virtualized. That means the Smartphones
E-UTRAN
conversion into software of MME (vMME)
(VM-n) (VM-m) (VM-o)
Server
Figure 1
2. LTE function-based virtualization: In this
Orchestration
model, LTE procedures and contexts (e.g.
Attach, Bearer Context) are virtualized User
Attach Bearers Plane
and LTE interfaces built around them.
Smartphones
E-UTRAN (VM-n) (VM-m) (VM-o)
Tablets Hypervisor
Server
Figure 2
3. Black-box virtualization:
This model treats the entire virtualized Orchestration
EPC as a black box with a set of external
E2E Virtual Evolved Packet Core
interfaces (e.g. S1, SGi, S6a). Smartphones
E-UTRAN (VM-n)
Tablets Hypervisor
Server Figure 3
• Lower hardware costs: Virtualization enables EPC functions to run on commercial off-the-shelf (COTS) servers,
simplifying the purchasing process and reducing supplier bargaining power
• Flexible automatic scaling: Virtual networks can flex in real time to meet demand. The orchestration management
system enables the mobile carrier to quickly optimize network configuration and topology through templates and
recipes in a few simple steps
• Fast development cycle: Virtualized functions can be easily upgraded and deployed in the network, enabling
carriers to offer more services, faster
• High Availability based on software: Automation of the vEPC with cloud orchestration and SDN increases
efficiency.
• Lower licensing costs: Most virtual infrastructure and functions will have a unified software platform. Although
vendors will initially propose proprietary solutions, the industry will favor open and expandable environments in
the long run
• Signaling load reduction: Communication between virtual nodes, especially signaling, can be reduced by
applying the proper adjustment of node collocation and routing. Signaling virtual functions can scale to respond
to instant demands
• Lower operational costs: A vEPC running on a standard NFV infrastructure will deliver some operational
efficiencies through reduced network costs and simplified operations
• Ease of supporting multiple regions (domain slicing) and carriers: Multiple carriers’ vEPC networks can be
configured on the same NFV infrastructure. This functionality also provides carriers with the flexibility to more
closely match core capacity to service demand
• Instant deployment and portability: The virtualized EPC can be used to quickly offload services (e.g. M2M),
handle temporary peaks of growth (e.g. social events), and deploy disaster recovery
3
How to Validate and Ensure a Successful Virtual EPC Deployment
A Complete, 4-Step Validation Process for Mobile Carriers Upgrading to a vEPC
This analysis shows how virtualizing the mobile core can be a highly cost-effective solution when planning the growth and
expansion of the network. However, carriers should not underestimate the challenges associated with deploying a virtual EPC
(vEPC), especially if the aim is to complete the move within a short timeframe.
Failure to acknowledge and address the challenges may lead to wasted investment, as the resulting vEPC does not deliver the
hoped-for performance increases, efficiency gains, or cost reductions.
With this in mind, the next section of this paper sets out a robust, four-step process for thoroughly testing the performance and
resilience of a vEPC prior to go-live.
In migrating the packet core to a vEPC, carriers will encounter a number of technical challenges. These can be grouped into two
broad categories: vEPC LTE functional challenges and virtual environment challenges. We examine each of these groups below.
www.spirent.com
Virtual Environment Challenges
The virtual environment must have certain characteristics in order to meet the challenges set out above. Below is a list of
minimal elements that the carrier must ensure are present in the infrastructure that will be supporting the virtual EPC, seen as an
application or VNF from the cloud point of view:
• Virtual infrastructure performance: Selection of the correct servers and hypervisor will be critical. These should
be optimized in terms of CPU usage, memory consumption and storage
• Virtual machine design: Each VM should provide maximum performance with minimal footprint. A key
consideration will be to minimize the “noisy-neighbor” effect, in which VMs compete for resources within the
same data center
• Orchestration: Carriers should design orchestration policies that successfully govern the lifecycle of each
virtualized network function (VNF)
• Automatic network scaling/sizing: Scaling and sizing should respond automatically to network demands, with
minimal transition times
• VM migration: The carrier must be able to increase network resiliency by implementing VM migration under
problematic network conditions
• Visibility: There must be high visibility of events occurring inside and outside the virtual environment
• Inter-working: The vEPC must be able to handle inter-working within data centers to maintain high availability
• Network optimization: The carrier must define network topologies that optimize network access, service and
resiliency
• Security: The infrastructure should be adequately protected from attacks
• Open environment: The network should leverage industry-standard APIs to interact with independent compute,
storage and networking hardware, hypervisors and cloud orchestration systems
This approach calls for the creation of a ‘validation ladder’, composed of four crucial steps – each of which encompasses a set of
mobile core key performance indicators. The building process can go up and down the ladder based on the results obtained at
each step (see figure 4).
Virtual
Infrastructure Step 1. Virtual Infrastructure Validation
Validation KPI
Degradation
Figure 4
5
How to Validate and Ensure a Successful Virtual EPC Deployment
A Complete, 4-Step Validation Process for Mobile Carriers Upgrading to a vEPC
The decision to move up or down the ladder will be taken as a result of analyzing two types of criteria:
1. Mobility event criteria: Defined as the success or failure to meet specific KPIs in terms of mobile events (e.g. % of failed
attempts, % of dropped sessions, etc.)
2. NFVI criteria: Defined as the success or failure to meet specific mobility KPIs without exceeding NFVI resource consumption
thresholds (e.g. CPU utilization, CPU max spike, etc.)
The next section sets out in detail the four evaluation steps on the validation ladder. For simplicity of illustration, we have chosen
to use a Virtual Node-based architecture as the reference System Under Test (SUT). This type of virtualization is the one most
commonly chosen by the carriers that Spirent works with, and is likely to emerge as the industry-standard approach to virtualizing
the EPC.
As discussed in the previous section, a virtualized infrastructure presents many validation challenges, as it is delivered through an
architecture consisting of many hardware and software layers. A typical virtual environment is depicted in Figure 5.
MANO
OSS/BSS NFVO
Os-Ma
Vn-Nf
NFVI
Virtualization Layer
VIM
Nf-Vi
Compute Storage Network
The infrastructure starts with commodity hardware—a standard, high-volume platform (switch, server, or storage)—and a
virtualization layer (hypervisor or container).
This combination of platform and virtualization layer is referred to as the NFV Infrastructure (NFVI) and it is managed by a
virtual infrastructure manager (VIM) such as OpenStack.
Virtual network functions (VNFs) run on top of this NFVI, and are managed by a VNF Manager, which is responsible for VNF
lifecycle management. Applications then access the VNFs transparently.
The NFV Orchestrator (NFVO) is responsible for on-boarding of network services and VNFs, service lifecycle management, and
other global resource management tasks.
www.spirent.com
Benchmarking the virtual infrastructure requires thorough analysis of the elements that compose it. The following documents
describe a set of methodologies which result in a detailed vision of the chosen virtual infrastructure:
[3.] Testing Methodologies for Validating NFV Environments, Spirent White Paper, October 2013
• Server/Virtualization Layer Performance Benchmarking: Detailing results around control plane and data plane
performance of the physical server and the virtualization layer
• Management and Orchestration: Focusing on VNF lifecycle testing; analysis of generic VNF on-boarding,
instantiation, scaling/updating, and termination
Using this vision as a guide, carriers can use key performance indicators (e.g. # of CPUs needed, CPU utilization, memory
utilization) as input for an optimization process that will alleviate any performance and reliability concerns resulting from
introducing virtualization.
The aim of the optimization process should be to obtain the most suitable combination of:
That entails isolating vMME, vSGW/vPGW and vPCRF, and conducting a set of tests to obtain capacity and performance metrics
for each node under varying network access conditions.
The results of these tests will provide the initial footprint for the network KPIs that will be used as reference for the optimization of
the final end-to-end network deployment.
Starting with a single instance of vMME (see figure), the validation should aim to determine the VNF’s maximum supported
subscribers and signaling events per second handled, while ensuring that the VNF adheres to its maximum allowed utilization of
NFVI resources (e.g. CPU, memory).
7
How to Validate and Ensure a Successful Virtual EPC Deployment
A Complete, 4-Step Validation Process for Mobile Carriers Upgrading to a vEPC
Orchestration
PCRF
VM VM MME VM VM
Spirent Landslide Spirent Landslide Spirent Landslide Spirent Landslide
Virtual Manager vRAN vNF vSGW/vPGW/vPCRF vIMS
Hypervisor
Server
Figure 6
As mentioned before, the first test should be to demonstrate that the vMME complies with the 3GPP standards in the S1-MME,
S1-U, S6 and S11 interfaces. Next, LTE Key Performance Indicators should be analyzed under several E-UTRAN access conditions
or call models. At a minimum, these tests should involve:
• Carrier BH Call Model: A combination of capacity and performance events following a specific traffic pattern
designed to represent the Busy Hour call model.
• Bearer performance
Similarly to the vMME, the gateway nodes (vSGW, vPGW or Combo vS/P-GW) must be capable of handling both control and user
planes simultaneously in interfaces such as S11 and S5/S8.
In today’s purpose-built gateways, dedicated resources within the nodes are allocated to these planes in order to guarantee
optimal treatment of user and control traffic. However, in the context of NFV, the VNFs and VNFCs may have to compete for NFVI
resources. A proper implementation of a vGW NF must ensure continuous and sustained QoS in the user plane regardless of
events occurring in the control plane, and vice versa.
www.spirent.com
Virtual node validation-node isolation
Virtual-gateway validation - “wrap-around” concept
Orchestration
PCRF
VM VM SGW VM VM
Spirent Landslide Spirent Landslide Spirent Landslide Spirent Landslide
Virtual Manager vRAN/vMME vNF vPGW/vPCRF vIMS
Hypervisor
Server
Figure 7
As with the vMME, the first test should be to demonstrate that the vGW complies with 3GPP standards in interfaces such as the
S11, S1-U and S5/S8. LTE Key Performance Indicators should then be analyzed under several network conditions or call models:
• Control Plane Benchmark: Capacity and performance test applied to the vGW control plane.
• User Plane Benchmark: High-throughput test based on an iMIX of small, medium and large packets
• Control and User Plane Decoupling tests: Tests that combine steady conditions with bursts of traffic in either the
control plane or the user plane. The goal will be to evaluate the correct allocation of NFVI resources between the
control plane and user plane in a vEPC Gateway
• Carrier BH Call Model: A combination of the previous tests following a specific traffic pattern designed to
represent the Busy Hour call model as seen in the carrier’s live network
For each of these conditions, KPIs of interest might include:
• Packets lost
9
How to Validate and Ensure a Successful Virtual EPC Deployment
A Complete, 4-Step Validation Process for Mobile Carriers Upgrading to a vEPC
The different test topologies should progressively move from the vMME-vSGW architecture to a final end-to-end topology where
all virtual nodes are active and connected (see Figure 8).
Orchestration
PCRF
Hypervisor
Server
Figure 8
If, in any topology, the SUT shows a degradation of the KPIs, the communication between VMs should be analyzed, and, if
necessary, a revision of steps 1 and/or 2 should be undertaken. Particular attention should be paid to any tests relating to VM
design, noisy neighbor, routing and control-plane/user-plane decoupling.
In this step, carrier Busy Hour call modeling testing becomes critical, as all elements of the network are stressed to create a
realistic representation of how the network will behave in production.
www.spirent.com
Step 4: Conduct End-to-End Service Validation
The final step in the ladder is to guarantee the optimal end-user experience of all the services using the virtual EPC.
This entails the deep analysis of each of the services, both individually and in combination with each other in the form of service
chains sustained by the NFVI and the vEPC as a whole.
For instance, a starting point would be to identify the services most likely to be run, such as VoLTE , IoT, Data Traffic handling
Service Chains or Hybrid Offloading, and to perform a two-plane, end-to-end analysis.
The first plane will be a test of the service isolated from the virtual infrastructure as shown in the VoLTE example below. Figure 9
shows a combination of emulated and real UEs connecting to a virtualized IMS and using a carrier-defined service chain element,
such as a Traffic Detection Function, to prioritize VoLTE traffic. This setup will use a full emulated infrastructure to further isolate
the VoLTE service chain from external interference sources.
Orchestration
QoE VoLTE Analysis with Prioritization
VM VM VM VM VM TDF VM IMS VM
Spirent Landslide Spirent Landslide Spirent Landslide Spirent Landslide Spirent Landslide Spirent Landslide Spirent Landslide
Virtual Manager vRAN vMME vSGW vPGW vNF vPCRF vIMS Internet Emulation
Hypervisor
Server
Figure 9
The tests to be conducted should focus on and analyze only the Key Performance Indicators relating to the service:
• Service Control Plane Benchmark: A capacity and performance test applied to the control plane of the service.
In the case of VoLTE, this will mean obtaining key measurements around the setup of SIP sessions; the creation of
dedicated bearers as indicated via the Rx interface; and the monitoring of procedure time with and without traffic
detection function enabled
• Service User Plane Benchmark: A high-throughput test. In the case of VoLTE, the analysis will focus on metrics
around RTP streams such as packet loss, jitter, latency and PoLQA scores with and without traffic detection
function enabled
• Service BH Call Model: A combination of previous tests following a specific traffic pattern designed to represent
usage of the service by several profiles of subscribers. In the case of VoLTE, such profiles will range from
subscribers that mostly use messaging and chat, to subscribers connected to video conferences, such as business
users
The second plane will be a test of the service in the network as a full end-to-end topology. Here, all of the elements are virtualized
nodes, and the functions will be subjected to real world traffic conditions.
The figure below shows the topology for the VoLTE case.
11
How to Validate and Ensure a Successful Virtual EPC Deployment
A Complete, 4-Step Validation Process for Mobile Carriers Upgrading to a vEPC
Orchestration
Hypervisor
Server
Figure 10
To validate the service in the network, the tests conducted in the isolated service validation should be repeated, but with the
addition of the carrier’s live call model used in Steps 1-3.
Such a model should use metrics obtained during the Busy Hour of the production network as input for the development of an
emulated call model. Some of the data measured could be:
Sub-call
Model 2
Sub-call TAU/RAU
Model 3
Figure 11
www.spirent.com
For each of these conditions, a combination of Network and Service KPIs should be analyzed. KPIs of interest might include:
• Network KPIs: Attach time, bearer creation/deletion time, bearers per second (applies to bearer activation, QoS
Change, HO), total aggregate traffic per APN, average packets per second processed, packets lost, average
packet latency and jitter, etc.
• Service KPIs (VoLTE example): Average Packet Size, Voice PPS, Throughput, MOS-Avg, MOS-Maximum
Once all of the tests and re-tests in the deployment validation ladder are complete, the end result is a fully functioning virtualized
mobile network in which every single component has been optimized for maximum service quality and resilience.
Equipped with this input, carriers can move on to confidently tackle the challenge of maintaining the KPIs as the Virtual EPC goes
live.
Figure 12
• Going-Live VNF Validation: As the operations team plans to deploy and/or upgrade a VNF in the production
network, that VNF should be tested with synthetic traffic prior to being connected into the live service chain. This
completely automated process provides a final validation to the orchestration system that the VNF is working all
the way up to the service layer.
13
How to Validate and Ensure a Successful Virtual EPC Deployment
A Complete, 4-Step Validation Process for Mobile Carriers Upgrading to a vEPC
vNF Turn-Up–MME
Orchestrator or
Manual Launch
vMME
Created by vMME
Orchestrator
vMME
Ready for vMME
Deployment
Live Network
Figure 13
• Pre-emptive Health Check: This methodology provides KPIs on the health of the live network. It consists of low-
scale tests mixed with live traffic to monitor the behavior of the production network at specific moments in time.
These tests inject virtual handsets into the network and run at scheduled intervals to obtain detailed KPIs on
service quality and responsiveness of each unique service chain.
• Degradation and Faults Isolation: If an issue is detected in any service chain during the pre-emptive health
checks or during monitoring of the live network, test VNFs are inserted into the service chain to isolate the
offending VNF.
This step is automated by the orchestration system as a policy, as the results of the analysis show whether the VNF is working as
designed, but it is oversubscribed (in which case the orchestration should dynamically scale this function), or the VNF is faulty and
a patch should be deployed.
Operations should apply the necessary changes via the ‘going-live’ methodology.
www.spirent.com
Using Spirent’s Landslide™ to Validate a Successful vEPC Deployment
All of the steps in the vEPC deployment validation ladder outlined in this document can be conducted with Spirent’s Landslide™
software.
Landslide enables service providers and equipment vendors to test the functionality, performance and scale of their networks
and nodes. Its highly-configurable end-points emulate real-world control and data plane traffic produced by millions of
subscribers moving through LTE, GSM, UMTS, eHRPD and Wi-Fi networks and consuming mobile broadband services (e.g. voice,
video conference, IoT, gaming, social).
Spirent Landslide supports unique testing methodologies that combine sets of isolated node and end-to-end test cases. As such,
it offers service providers and equipment vendors a fully controlled test environment to:
• Analyze impact of Busy Hour Call Modeling that includes mobility to/from LTE, GSM, UMTS, eHRPD and Wi-Fi
networks
1. Support for all 3GPP and 3GPP2 protocols and interfaces, with complete stateful behavior
2. Ability to seamlessly conduct all tests inside the virtual infrastructure, allowing full visibility of the network
under test. Landslide can run on all major hypervisors and supports a wide range of virtual network interfaces.
Test VNFs can be easily orchestrated via REST-API for a full embedding in the carrier’s data center. In doing
so, service providers and carriers can easily control Spirent Landslide’s test functions from their proprietary
environment without needing deep knowledge of the test tool.
3. Test cases covering the full spectrum of call models to validate virtual EPC nodes. Models include capacity,
performance, service modeling and carrier Busy Hour call modeling at high-scale, as specified in the 4-step
validation ladder.
4. Comprehensive set of core node emulators, delivered in the form of Virtual Network functions to facilitate
testbed completion during node isolation, progressive end-to-end and service validation tests.
15
How to Validate and Ensure a Successful Virtual EPC Deployment
A Complete, 4-Step Validation Process for Mobile Carriers Upgrading to a vEPC
Rev B | 08/18