Trend-Based Networking Driven by Big Data Telemetry for Sdn and Traditional Networks

International Journal of Next-Generation Networks (IJNGN) Vol.11, No.1, March 2019
DOI : 10.5121/ijngn.2019.11101 1
TREND-BASED NETWORKING DRIVEN BY BIG
DATA TELEMETRY FOR SDN AND TRADITIONAL
NETWORKS
Ankur Jain1
, Arohi Gupta2
, Ashutosh Gupta3
, Dewang Gedia4
, Leidy Pérez
5
, Levi
Perigo6
, Rahil Gandotra7
and Sanjay Murthy8
Interdisciplinary Telecom Program, University of Colorado Boulder, USA
ABSTRACT
Organizations face a challenge of accurately analyzing network data and providing automated action
based on the observed trend. This trend-based analytics is beneficial to minimize the downtime and
improve the performance of the network services, but organizations use different network management
tools to understand and visualize the network traffic with limited abilities to dynamically optimize the
network. This research focuses on the development of an intelligent system that leverages big data
telemetry analysis in Platform for Network Data Analytics (PNDA) to enable comprehensive trend-
based networking decisions. The results include a graphical user interface (GUI) done via a web
application for effortless management of all subsystems, and the system and application developed in
this research demonstrate the true potential for a scalable system capable of effectively benchmarking
the network to set the expected behavior for comparison and trend analysis. Moreover, this research
provides a proof of concept of how trend analysis results are actioned in both a traditional network and
a software-defined network (SDN) to achieve dynamic, automated load balancing.
KEYWORDS
Ansible, Big Data, Collectd, Jupyter, Kafka, Logstash, Network Analysis, PNDA, Python, OpenFlow, Ryu,
SNMP, Software-defined Networking, Trend Analysis
1. INTRODUCTION
In the early-to-mid 1990s the Internet’s growth changed the rules and due process in the daily
lives of users all around the globe. What started as email for scientists and early file transfer
systems, is now what moves every fiber of society regardless of culture, expertise or even direct
exposure to connectivity [1]. This vast global adoption led to a shift in paradigm from
manageable quantities of network data to a massive volume of traffic. Automation and
orchestration applications such as virtual load balancers have become critical to cloud and data
center control systems [2, 13, 15].
According to Cisco’s figures for global IP growth for mobile and fixed networks, presented in
their Visual Networking Index (VNI) Forecast report [3], the number of global Internet users will
reach 4.6 billion by 2021 with 27.1 billion network devices, generating over 3.3 zettabytes
of IP traffic per year. Therefore, the Internet has entered big data telemetry territory, kicking off,
as Cisco coined it first, The Zettabyte Era [4].
The main challenges that network managers are facing are scalability, analytics and actioning of
results. Numerous Network Management System (NMS) tools have been developed with the goal

2
of making data reveal the status of large networks. However, there is little automated action on
the analytics results, which often include metrics on observable trends in network operations [5,
16].
The solution presented in this paper is an intelligent system that leverages the scalability and
analytics capabilities of the Platform for Network Data Analytics (PNDA), to then enable
automated networking decisions to be executed in the underlying network topologies.
The remainder of the paper is organized as follows: Section II provides a review of the existing
body of knowledge, state of the art applications, and how our scheme extends it. Sections III and
IV describe the methodology and results of our experiment respectively. Section V concludes our
research and addresses scope for future enhancements.
2. RELATED WORK
When it comes to network analytics, there have been related efforts around providing a network
manager with a global view of the performance of their network. The common output is a visual
representation of data via dashboards and an alarm functionality for unexpected changes in
performance. The end goal being passive monitoring that serves as an aid to network operations.
However, the demand for functionality besides monitoring is overwhelming [6, 18, 19, 20, 21].
Enterprise Management Associates (EMA) did a research study titled “Network Management
Megatrends 2016: Managing Networks in the Age of the Internet of Things, Hybrid Clouds and
Advanced Network Analytics” [7]. Their premise was in this big data era, enterprise network
managers are starting to have “higher aspirations” for the applications of the telemetry data
collected from their networks. The results showed that, at the time of the study, advanced
analytics tools have a deep impact in not only network operations, but in business applications.
Network security monitoring resulted in the main advantage sought after with 38% of their data
set, closely followed by network optimization as the second most popular use case with 32% of
support. Their work not only backs up the multiple use cases for the research in this study, but
also the increasing demand in network optimization driven by big data, specifically, big data
telemetry. Big data telemetry defined as the robust collection of enormous quantities of network
data and its aggregation onto one centralized big data analytics platform for analysis, which is the
key decision driver of the architecture designed for the system presented in this paper.
Furthermore, there are previous efforts aimed at providing a solution for a network topology that
includes both traditional and software-defined networking (SDN) topologies [13, 15].
Additionally, network engineering practices face a major challenge of network optimization in
this hybrid network scenario [8, 16].
One of these initiatives was brought forward by the Polytechnic School of Engineering from New
York University. Their emphasis was to develop a congestion-aware single link failure recovery
system for hybrid SDN [9], in order to achieve fast failover recovery and load balancing post
recovery. The way they achieved this was by developing a heuristic algorithm that utilizes SDN
devices in the recovery path that have a global view of the network to formulate recovery routes.
The key difference from their work to the solution presented in this research is that they solved
for a hybrid network, one with both traditional and SDN devices interfaced together, versus
having a mixed solution by having the two network topologies be independent of each other, like
the one this research focuses on.
In terms of automated load balancing, a team from the Ryazan State Radio Engineering
University worked on developing an improved model of multipath adaptive routing in computer
networks with a focus on load balancing capabilities [10]. Their approach on load balancing was

3
heavily focused on the indicator of jitter optimization to select an optimal route. Their work
resulted in an algorithm capable of acting as a baseline for new routing protocols, in contrast with
the research for this study which presents a holistic user-oriented solution by integrating multiple
subsystems to achieve a trend-based load balancing platform equipped with big data analytics and
a Graphical User Interface (GUI) for ease of management.
In the article “A Dynamic Bandwidth Allocation Algorithm in Mobile Networks with Big Data of
Users and Networks” [11] the authors utilize big data telemetry collected from the network to
“cluster users by analyzing their closeness, both geographical and social, such that users in the
same cluster share a wireless channel for downloading contents from the base station.”
Effectively showing how leveraging big data for dynamic resource allocation optimizes network
management practices. However, the work done in this research differentiates itself from their
research by having the core of the decision-making capabilities designed around trend analysis
over time, creating an intelligent predictive system, versus real time resource allocation.
This research aims to answer the research question: “Can big data collection, indexing,
filtering, aggregation, benchmarking, and trend based analysis and actioning be service-
chained to collectively provide a centralized big data analytics platform for telemetry data
ingestion and analysis?” This research question was strategically divided into sub-problems to
answer the primary research question:
a. Can SNMP and OpenFlow be used to fetch network-wide telemetry data from traditional
and software-defined networks respectively and store this information in a centralized
database?
b. Can Logstash be used to filter and sort the data metrics to serve as input to PNDA’s data
streaming interface, Kafka? If so, can Jupyter Notebook streamline the data from Kafka
interface to OpenTSDB through PNDA’s network analysis tools?
c. Can statistical benchmarking be performed on the stored telemetry data by real-time
monitoring to identify trends and take appropriate action?
The novel contribution of this paper is to design and implement a network architecture and an
application that can be used together in a hybrid network to perform trend-based networking
through analytics and proactive network monitoring. Furthermore, this research developed a self-
sufficient system capable of making intelligent decisions through leveraging a powerful big data
network analytics tool to load balance traffic in both SDN and traditional networks based on
observable trends in link utilization. This research is beneficial to organizations because not only
does it analyze network traffic, but also dynamically takes a corrective action for the identified
network trend; thus, achieving trend-based, automated network optimization for better overall
network performance.
3. RESEARCH OVERVIEW
The design solution for this system is developed for scalability, analytics, and actioning of results.
All subsystems are carefully integrated to ensure telemetry collection is able to scale to massive
data sets, given all hardware specifications are adjusted with the increase in traffic expectancy.
The core component driving the design of all subsystems was PNDA, acting as the centralized big
data analytics platform for telemetry data ingestion and analysis. See Figure 1.

4
Fig. 1. Test Architecture
A. Can SNMP and OpenFlow be used to fetch network-wide telemetry data from traditional
and software-defined networks respectively to store it in a centralized database?
The first step in the system is to perform the telemetry data collection from the underlying
network topologies. This was achieved by designing a network configuration that meets the
design requirements of SNMP data collection from traditional networks and switch statistics
retrieval from SDN. SNMP data is collected from the traditional network using a systems
statistics collection daemon (CollectD) to set frequency of data collection and tag traffic with the
host’s source IP. The traditional network topology can be seen in Figure 2, where SNMP agents
have been set up in each device.
Fig. 2. Traditional Network Topology
The SDN topology is shown in Figure 3. To collect switch statistics from it, it is necessary to use
a REST API and the OpenFlow protocol to fetch counters from each SDN switch via network
flows. To meet this requirement, the selection of the SDN controller was based on the best fit
REST API for the purpose, resulting in the implementation of the Ryu SDN controller. The
resulting network architecture design enables a scalable telemetry data collection subsystem for
both SDN and traditional networks.

5
Fig. 3. SDN Topology
B. Can Logstash be used to filter and sort the data metrics to serve as input to PNDA’s data
streaming interface, Kafka? If so, can Jupyter Notebook streamline the data from Kafka
interface to OpenTSDB through PNDA’s network analysis tools?
The second step the system follows is to filter all telemetry data and prepare it for the big data
analytics tool, PNDA. The design requirement was to have all raw data, collected from both
networks, be filtered and formatted for seamless ingestion into PNDA. The tool selected for this
development was Logstash, a data processing pipeline used to ingest data from different sources
into a common platform with the required format [12].
For the traditional network, this is done by taking the data coming from the network via a
CollectD configuration file written specifically for this project, and sending it to Logstash’s input
plugin. The SDN is capable of sending data directly into the same plugin without being filtered
through CollectD.
Logstash was configured to run the data through its filter plugin and sort through it to get only
specific metrics such as outPkts and outOctets. This data set is then sent to the output plugin,
which serializes and encodes it using a codec named ‘avro-codec’, a requirement to ensure
compatibility with the following subsystem. This sequence allows the system to send to PNDA’s
data streaming interface, Kafka, only valuable and properly formatted data, which in turn
streamlines the process and reduces the workload on the server [17].
Fig. 4. Logstash Indexing Instance

6
The third step built into the system’s workflow is the subsystem to aggregate all data coming
from Logstash to PNDA. At this point, all data collected from both SDN and the traditional
network is ready for PNDA’s ingestion. The main design requirement for this stage is to get data
from Kafka, the data streaming interface in charge of parsing through the data in real time, to
PNDA’s own Time Series Data Base, OpenTSDB. To do this, the main development was to
implement an original Jupyter notebook.
A Jupyter notebook is the format of python algorithm Kafka is capable of integrating with. It
contains all commands on what to do with input, data incoming from Logstash, and where to
store the output. This stage is coded to go from Kafka, through PNDA’s network analysis tools,
to finally store results in OpenTSDB and make them available via Impala, a standard SQL
interface used to query data from PNDA. See Figure 5.
Fig. 5. PNDA Cluster
This results in data being ran through PNDA’s analytics engine and getting ready to be utilized
for benchmarking and trend analysis in the following subsystems.
C. Can statistical benchmarking be performed on the stored telemetry data by real-time
monitoring to identify the trend and take appropriate action?
The fourth step for the integrated system to go through consists of leveraging PNDA’s analytics
to set a benchmark for both network configurations and to determine what behavior in the active
topologies constitutes a trend that needs action in time versus changes that are one time
occurrences. The development required for this sub-system was a GUI, see Figure 6, with two
main components: a Python back-end architecture and a bootstrap front-end development. The
front-end design, see Figure 7, is a comprehensive user-centric web application designed to
address network engineers’ needs of centralized resources to manage a system of this magnitude.
To avoid duplicating efforts, it also has a redirect tab to PNDA’s own dashboard for analytics, see
Figure 8.

7
Fig. 6. Graphical User Interface (GUI) block diagram
Fig. 7. GUI Front-End
Fig. 8. PNDA’s Dashboard

8
The back-end scheme is the engine that integrates all subsystems up to this step by performing
two global tasks, benchmarking and trend analysis. Benchmarking runs all commands to execute
subsystems one through three to then consume data analytics coming from PNDA and execute
the algorithm’s section developed to evaluate network behavior without intervention. This unit of
the code is set to query all telemetry data for an established time period and frequency, to then
run a statistical distribution of data and get the mean value of link utilization for the main output
ports in the system. It will also set a threshold of interface bandwidth to separate “normal traffic,”
from “high traffic,” which will also be used as a trend indicator. This process will repeat to reset
the benchmark with a user determined frequency. See Figure 9.
For trend analysis, the code was developed to identify link utilization trends that will be utilized
for load balancing. It takes the statistical values and threshold established during benchmarking to
find the standard deviation of new incoming traffic, all while monitoring the interface bandwidth
threshold. Once traffic crosses the threshold and deviates significantly from mean value of traffic,
the code will raise a flag to mark the trend, until traffic lowers back to below the threshold or
within regular values, and the change that needs to be addressed is established in a timeframe.
This process effectively enables the back-end to action a specific trend and make the right
decision to load balance the system during the right period of time.
Once data is collected, filtered, ran through big data analytics and submitted for benchmarking to
identify the trend, the system is ready for the call to action on the latest trend spotted by the back-
end, all while providing visibility in the front-end. The call to action of a trend-based networking
decision is executed by pushing the appropriate data structure to the correct devices in each
topology. For the traditional network, this is done via an open-source library that enables SSH to
routers, Netmiko. Which will send the appropriate route-map CLI commands to the key affected
router. The route-map contains the CLI commands necessary for traditional routers to associate a
lower local preference for the specific affected subnet and a higher local preference for the new
path, which determines where traffic is routed during the timeframe needed. For the SDN
topology, the same concept is used, with the difference of flows entries being pushed to the SDN
instead of route-maps.
.
Fig. 9. Benchmarking Code Flow

9
OpenFlow flow entries will determine the change in path for traffic during the timeframe
established. Once a decision has been made and pushed to the network, the user can choose to
have the decision be effective for a period of time and then revert it to reevaluate the status over
time in their network. The logic followed by the code is detailed in Figure 10.
Fig. 10. Trend-Based Networking Decision Block Diagram
When all steps are performed in an integrated, seamless manner, the end result is a trend-based
networking system that utilizes powerful tools, such as PNDA for big data telemetry analysis, to
provide the user with the most powerful interpretation and usage of their network data. Not only
does the user get all monitoring benefits from PNDA’s built in functionality, but also gets a GUI
on which to track trends and actions taken in the network to keep traffic running without
congestion or packet loss, with minimal human intervention.
4. RESULTS AND ANALYSIS
A. Physical Test Setup
Two servers were utilized to distribute all components of the system as seen in Figure 11. The
hardware and software specifications are broken down in Table 1 and Table 2 respectively.
Multiple levels of virtualization and different types of servers were used to complete this physical
setup. The Server 1 is a bare metal server running Ubuntu 16.01 as its Virtualization Layer 0. On
this Ubuntu Server is where VirtualBox is set up, making Virtualization Layer 1. Then,
VirtualBox has Ubuntu 14.04 installed as Virtualization Layer 2, which contains the PNDA VM
and the GUI as its logical components. Server 2 is dedicated to network virtualized topologies.
This bare metal server runs VMware Esxi 6.5 as its Virtualization Layer 0 and two Virtual
Machines (VMs) running Ubuntu 16.01, Virtualization Layer 1, for each network design.

10
Fig. 11. Physical setup
Table I. Hardware Specifications
Device
Hardware Specifications
Make Model CPU Memory
12 CPUs x
PowerEdge
Intel(R)
Servers Dell Inc Xeon(R) CPU 32 GB
R430 E5-2603 v3 @
1.60GHz
Table II. Software Specifications

11
The first VM has the traditional network topology generated using GNS3. The second VM has
the SDN components, in which the Ryu controller and Mininet are set up to establish the
topology. These two VMs are connected to each other using a virtual switch, and further
connected to a router which allows connectivity to the Internet using port forwarding.
B. Proof of Concept
Establish Benchmark
In order to determine the mean value of link utilization of output ports, it was necessary to first
perform telemetry data collection and analysis of the network in a period of 72 hours, in order to
get the baseline behavior. This was achieved by implementing the sub-system integration from
step 1 through step 4 of the workflow described in section III A, Figure 1.
First, traffic was generated using a network performance tool called iPerf 3. The frequency was
set to every 1 hour, 24 samples per day, and 72 samples in total. The network topologies were set
up as displayed in Figures 2 and 3.
The assumption made for traffic generation was that the highest activity hours are between
business hours, 8 AM to 5 PM, with activity decreasing significantly outside of that time frame.
Then for step two, as traffic was generated, it was aggregated by CollectD to Logshtash’s input
plugin for the traditional network and directly from the RYU controller REST API. Logstash ran
the data through its filter plug in to relay inPkts, outPkts, inOctets, outOctets, interface name and
time stamps to PNDA’s Kafka interface using the pnda-avro plugin. See Figure 12 for a sample
of the output coming from Logstash’s output plugin.
Fig. 12. Output from Logstash’s Output Plugin
Now the system is at step three and traffic has arrived to PNDA. Once the Kafka interface detects
the data input, a Jupyter query takes the telemetry data and sends it to the OpenTSDB database
for storage. A snippet of data stored in OpenTSDB can be seen in Figure 13. This shows how
once telemetry data gets to this stage, it is already separated by type. For this snippet, the display
is a snapshot in time for outOctets.

12
Fig. 13. Snippet of outPkts in OpenTSDB
Once telemetry data is analyzed and separated by type of input or output, comes the fourth step in
the systems sequence, which consists of the back-end application retrieving data logs from
Impala to set the benchmark for trend analysis. The results of the benchmarking in the back-end
are shown in Figure 14.
Fig. 14. Benchmarking performed in the back-end
From the graph, it can be observed how the data samples were collected for three (3) days with a
frequency of once per hour. A mean value was calculated for every hour separately. This mean
value is used as the benchmark for that particular hour and will be used in the trend based
analysis in future. These results enable the second round of testing, which will entail taking this
benchmark to identify trends and send automated corrective action back to the topologies,
fulfilling the last step in the system’s workflow.
5. CONCLUSIONS AND FUTURE WORK
In this paper, all steps to achieve the full integration of the trend-based networking system driven
by big data telemetry for SDN and traditional networks have been described. It has been proven
that benchmarking the network using all the subsystems built in the design solution is possible
and effective. Moreover, the prospect of the tests conducted in this research can further be
calibrated based on the desired use-case for automated load balancing and trend-based
networking. The contributions of this research relies on the algorithms carefully developed to run
subsystems at the top of their performance, while integrating seamlessly to create a fully
integrated system. The user gets both the benefits of monitoring all the data PNDA is analyzing
via PNDA’s own dashboard, plus an automated trend-based load balancing system administered
via the system’ own GUI with little human intervention needed.

13
This design sets the foundation for the next generation of network management system tools. The
number of applications that can be built on top of this work is only limited by what application
the user envisions in their network. Some of the next steps and applications that could be
implemented are:
A. With the benchmark set in the performed test along with the mean value and link utilization
threshold, the next step is to collect data again over a span of seven days, this time picked
tentatively as a generic number for testing purposes. The frequency of collection will be the same
used in benchmarking, once an hour, to keep compatibility between data sets. Once the telemetry
data set has gone through all the same subsystems that the benchmarking test used, it will be time
to identify the trend by analyzing the standard deviation of the data sample for every hour against
its counterpart in the benchmark data set, following the system logic detailed in Figure 10.
B. Expand the scope of network metrics used for trend-based decisions: link utilization was used
as a proof of concept. This network can be scaled to include any network metric compatible with
the data collection system, such as CPU utilization, bandwidth, and latency.
C. Machine Learning for Network Trends: leveraging PNDA’s big data capabilities, it is possible
to keep a comprehensive archive of all trends identified and have a machine learning algorithm
running in the back end to start predicting network behavior before the trend even occurs.
D. Evolve from load balancing via prefix to load balancing via multipath: our current solution
implements a prefix-centric load balancing approach. If this is taken one step further, multipath
BGP can be implemented for further load balancing capabilities.
E. Connecting traditional and SDN: while this solution is oriented towards SDN and traditional
networks operating independently, it can be developed to support both networks communicating
enabling all hosts pertaining to both networks to achieve end-to-end connectivity.
REFERENCES
[1] Castells, M. (2018). The Impact of the Internet on Society: A Global Perspective - OpenMind.
[online] OpenMind. Available at: https://www.bbvaopenmind.com/en/article/the-impact-of-the-
internet-on-society-a-global-perspective/?fullscreen=true.
[2] B. Kleyman, “Five Critical Layers of Next-Gen Data Center Automation and Orchestration,” Data
Center Knowledge, 17-Aug-2015. [Online]. Available:
http://www.datacenterknowledge.com/archives/2015/08/17/five-critical-layers-of-next-gen-data-
center-automation-and-orchestration.
[3] Cisco Visual Networking Index: Forecast and Methodology, 2016–2021,” Cisco, 16-Sep-2017.
Available [online]: https://www.cisco.com/c/en/us/solutions/collateral/service-provider/visual-
networking-index-vni/complete-white-paper-c11-
481360.html.ION=Cisco+link&REFERRING_SITE=Cisco+page&C%20REATIVE=CMP+infog+to
+VNI+WP.
[4] “The Zettabyte Era: Trends and Analysis,” Cisco, 01-Aug-2017.[Online]. Available:
https://www.cisco.com/c/en/us/solutions/collateral/service-provider/visual-networking-index-vni/vni-
hyperconnectivity-wp.html.
[5] J. Loeb, “Network Monitoring: Best Practices in the Face of IT Complexity,” The Data Center
Journal, 04-May-2016. [Online]. Available:http://www.datacenterjournal.com/network-monitoring-
best-practices-face-complexity/.

14
[6] L. Wolfaardt, “How an effective Network Management System benefits your business,” Iris Network
Systems, 14-Apr-2016. [Online]. Available:https://www.irisns.com/how-an-effective-network-
management-system-benefits-your-business/.
[7] S. McGillicuddy, “Network Management Megatrends 2016: Managing Networks in the Era of the
Internet of Things, Hybrid Clouds and Advanced Network Analytics,” Apr-2016. [Online].
Available:https://www.riverbed.com/document/fpo/Riverbed-EMA-NetworkMgmtMegatrends-2016-
RR-SUMMARY_CS1.pdf.
[8] S. Vissicchio, L. Vanbever, and O. Bonaventure, “Opportunities and Research Challenges of Hybrid
Software DefinedNetworks,”2014.[Online].Available:
http://vanbever.eu/pdfs/vanbever_hybrid_sdn_ccr_2014.pdf.
[9] C. Y. Chu, K. Xi, M. Luo and H. J. Chao, "Congestion-aware single link failure recovery in hybrid
SDN networks," 2015 IEEE Conference on Computer Communications (INFOCOM), Kowloon,
2015, pp. 1086-1094.
[10] V. P.Koryachko,D. A.Perepelkin,and V.SByshov,“Development and Research of Improved Model of
Multipath Adaptive Routing in Computer Networks with Load Balancing,”31-Oct-2016.
[Online].Available:https://link-springer-
com.colorado.idm.oclc.org/content/pdf/10.3103/S0146411617010047.pdf.
[11] B. Fan, S. Leng and K. Yang, "A dynamic bandwidth allocation algorithm in mobile networks with
big data of users and networks," in IEEE Network, vol. 30, no. 1, pp. 6-10, January-February 2016.
[12] “Logstash” Open Source Search & Analytics • Elasticsearch, 18-Apr-2017. [Online]. Available:
https://www.elastic.co/products/logstash
[13] D. Gedia, and L. Perigo, “NetO-App: A Network Orchestration Application for Centralized Network
Management in Small Business Networks” in 4th International Conference on Computer Science,
Engineering and Information Technology (CSITY-2018), Sydney, Australia, July, 2018, pp. 61-72,
DOI: 10.5121/csit.2018.81106.
[14] Netmiko Library [Online]. Available: https://pynet.twb-tech.com/blog/automation/netmiko.html
[15] D. Gedia, and L. Perigo, “A Centralized Network Management Application for Academia and Small
Business Networks” in Information Technology in Industry (ITII). Aug, 2018, pp. 1-10, Vol. 6, No. 3.
[16] R. Gandotra, and L. Perigo, “SDNMA: A Software-defined, Dynamic Network Manipulation
Application to Enhance BGP Functionality” in 20th IEEE International Conference on High
Performance Computing and Communications (HPCC-2018), June, 2018.
[17] Kafka Messaging Bus [Online]. Available: https://kafka.apache.org.
[18] N. Elgendy, and A. Elragal, “Big Data Analytics: A Literature Review Paper” in Advances in Data
Mining. Applications and Theoretical Aspects: 14th Industrial Conference, ICDM 2014, July, 2014,
pp.214-227, DOI 10.1007/978-3-319-08976-8_16.
[19] M. Hadi, et. al., “Big data analytics for wireless and wired network design: A survey” in Science
Direct, Feb., 2018, Vol. 32, pp. 180.199.
[20] K. Sundaresan, et. al., “Big data analytics approach for network core and edge applications” in 2016
IEEE Aerospace Conference, 2016, pp. 1-10.
[21] D. Gedia, and L. Perigo, “Performance Evaluation of SDN-VNF in Virtual Machine and Container”,
in 2018 IEEE Conference on Network Function Virtualization and Software Defined Networks (NFV-
SDN), Verona, Italy. (2018).

15
AUTHORS
Ankur Jain is a graduate student from the University of Colorado Boulder with
a major in Network Engineering. He currently works as an Associate Systems
Engineer at Juniper Networks.
Arohi Gupta is a graduate student from the Interdisciplinary Telecom Program,
University of Colorado Boulder. She received her bachelor’s degree in
Electronics and Telecommunications Engineering, and currently works as
Software Engineer 3 at Juniper Networks.
Ashutosh Gupta is a graduate student from the University of Colorado Boulder
with a major in Network Engineering. He received his bachelor’s degree in
Electronics and Telecommunications Engineering, and currently works as
SDN/NFV Software Engineer 3 at Juniper Networks.
Dewang Gedia is a Ph.D. candidate at the Interdisciplinary Telecom Program,
University of Colorado Boulder. He received his master’s degree in Network
Engineering, and has primary research focus in network functions virtualization
and software-defined networks domain.
Leidy Pérez is a graduate student from the Interdisciplinary Telecom Program,
University of Colorado Boulder. She received her bachelor’s degree in
Telecommunications Engineering, and currently works as Strategic Operations
Manager at Zayo Group.
Dr. Levi Perigo is a Scholar in Residence and Professor of Network Engineering
at the Interdisciplinary Telecom Program, University of Colorado Boulder. His
interests are in a variety of internetworking technologies such as network
automation, VoIP, IPv6, SDN/NFV, and next generation protocols. Currently, his
research focuses on implementation and best practices for network automation,
SDN, and NFV.
Rahil Gandotra is a Ph.D. student at the Interdisciplinary Telecom Program,
University of Colorado Boulder. He received his bachelor’s degree in
Telecommunications Engineering, and has primary research interests in next-
generation networking focusing on software-defined networking, network
functions virtualization, and energy-efficient networking.
Sanjay Murthy is a graduate student from the Interdisciplinary Telecom
Program, University of Colorado Boulder. He received his bachelor’s degree in
Electronics and Communications Engineering, and currently works as Production
Network Engineer at Facebook.

Trend-Based Networking Driven by Big Data Telemetry for Sdn and Traditional Networks

More Related Content

What's hot (12)

Similar to Trend-Based Networking Driven by Big Data Telemetry for Sdn and Traditional Networks (20)

More from josephjonse (20)

Recently uploaded (20)

Trend-Based Networking Driven by Big Data Telemetry for Sdn and Traditional Networks