Transcript Anna Maria Mandalari, ESR 13 UC3M

METRICS
ITN-Multi beneficiary
Mid Term review 18 September 2015
Agenda
•
18th September
•
9:00-10:00 Project overview [60min]
– Project presentation and education activity (summer schools, seminars) [25min]
– Project and WP status [20min]
– Placement planning [10min]
– Publications, submissions and dissemination [5 min]
•
10:00-12:30 Students presentations (1/2)
– 2h Student presentation (15’ each)
•
Lunch (12:30-13:30)
•
13:30-15:00 Student presentation (15’ each)
•
15:00-16:00 Meeting students with REA representative
Closing
STUDENT PRESENTATIONS
• Diana Andreea Popescu, ESR4 UCAM
• Supervisor: Andrew Moore
Background
• Master of Science in Computer Science at École
Polytechnique Fédérale de Lausanne, 2011-2013
• Bachelor of Science in Computer Science at University
Politehnica of Bucharest, 2007-2011
• Internships at CERN (master thesis and summer student)
and at Google in Site Reliability Engineering and in Platforms
Networking
Diana Andreea Popescu, ESR4 UCAM
4
Measurements in Data Centre Networks
• Research focus: investigate techniques and ways that enable
efficient network measurements in data centres using
software defined networking
• Analysed the fidelity of experiments in network emulation
environments
• Gained understanding of data center workloads by running
applications in the network emulation environment
• Identification of metrics needed in data centres and purposes
for network monitoring (debugging, discovering traffic
patterns used in job placement, etc.)
Diana Andreea Popescu, ESR4 UCAM
Measurements in Data Centre Networks
• Designing components for network monitoring in data centre
address its needs and that scale to its size
• Study data centre network traffic characteristics and propose
model
• Challenges: no network traces from data centres, as
companies do not publicly release them
Diana Andreea Popescu, ESR4 UCAM
6
Trainings and conferences
Training
• Datacenter Simulation Methodologies Tutorial organised as
part of the 47th Annual IEEE/ACM International Symposium
on Microarchitecture, December 2014, Cambridge
• ACM Symposium on SDN Research, 17-18 June 2015,
Santa Clara, California (travel grant)
• CoNext 2015 Student Workshop submission (work in
progress)
• To attend ACM IMC 2015, 28-30 October, Tokyo, Japan
(travel grant)
Diana Andreea Popescu, ESR4 UCAM
7
STUDENT PRESENTATIONS
• Zafar Gilani, ESR5 UCAM
• Supervisors: Jon Crowcroft
Student #2 – Zafar Gilani
• Social Media & Mobile Networks
– Attended Summer School 2015
– Work on
• what can we infer from online social media
• Paper submitted to hotnets
• About future network demand
– Started on road traffic location & cellular net load
• Passed 1st year viva…
– Now Looking at developing world:-
• Paper: epidemic search term&web load in Africa
• Accepted for IEEE CCNC
Gilani Background
• July 2013 Kungliga Tekniska Högskolan (KTH) & Universitat
Politècnica de Catalunya (UPC) European Master in Distributed
Computing (EMDC) (Hons) Final score: 8.34/10.0
• August 2009 NUST School of Electrical Engineering and Computer
Science Bachelor of Information Technology (Hons) Cumulative
GPA: 3.58/
Gilani Record
• IT Specialist at GfK SE, Nuremberg, Date: Oct 2013 to Sep 2014
• Research Intern at Telefonica Research, Barcelona, Date: Jan 2013 to Jul 2013
• CONFINE intern at Universitat Politècnica de Catalunya, Barcelona, Date: Apr 2012
to Jun 2012
• Research Assistant at NUST SEECS, Islamabad, Date: Jul 2011 to Aug 2011
• Example Publication
November 2013 - Finamore, A., Mellia, M., Gilani, Z., Papagiannaki,K., Grunenberger, Y.,
Erramilli, V. “Is There a Case for Mobile Phone Content Pre-staging?” , In CoNeXT 2013 Won
the best short paper award.
STUDENT PRESENTATIONS
• Miriam Marciel, ESR3 NEC Labs
• Supervisors: Saverio Niccolini/Mohamed Ahmed
Presentation


Bsc. and Msc. In Telematics Engineering from Universidad
Carlos III de Madrid (UC3M) in 2012 and 2013
Research Assistant at IMDEA Networks (Madrid) for 2 years
Miriam Marciel, ESR3 NEC Labs
My work in the project

WP2: Big Data Analysis for Advanced Network Management
& Operation
 Task 2.1: Performance optimizations of distributed stream
processing platforms

Task 2.2: Stream mining of content popularity lifetime
evolution and its applications

Supervisors:

NEC: Saverio Niccolini, Mohamed Ahmed

UC3M: Marcelo Bagnulo
Miriam Marciel, ESR3 NEC Labs
Research activities and results

Measurement and analysis of the online advertising
ecosystem:
- Measuring fraud in Video Content Portals (to be submitted to
PAM'16)
- Measuring value of the user on the Internet (current work)
- Studying targeting advertisement (Under review at PKDD
journal 12/15)
- The use of cookies (Under review at CCR Jan'16 edition)
Miriam Marciel, ESR3 NEC Labs
Motivation
Online advertising generated $49B in the US
Part 1: Understanding video ad fraud


Video ads generated $3.3B in 2014, only in the US, expected
to grow to $8B in 2016

23% of monetized views are fraudulent [1]
Part 2: Valuating users

Online advertisement is largely based on user targeting

How are users targeted?

How are users priced?
[1] ANA and White Ops, “The Bot Baseline: Fraud in Digital Advertising.”
https://www.ana.net/getfile/21853, December 2014
Miriam Marciel, ESR3 NEC Labs
Measurement tools


Automatic views generator

Selenium Webdriver

Modular

Isolation of videos

400 IPs addresses: Spain, Germany, Planetlab
Fetching Statistics from Video Portals


Public view counters and monetized counters
Performance metric
Rate of False Positives: fake view misclassified and counted in the
view counter
Miriam Marciel, ESR3 NEC Labs
Is there any detection system on the
public view counter in major portals?


YouTube is the only one detecting
Dailymotion seems to have a detection mechanism but for a high
volume of views
Miriam Marciel, ESR3 NEC Labs
Is different the detection system for
monetized and public counters?

Hypothesis: A stronger control should apply to monetized counters

Different systems for the public and monetized view counter

YouTube is more permissive with the monetized counter than the public view
counter
Miriam Marciel, ESR3 NEC Labs
Valuating the users


Collecting data from the suggested bids given to
advertisers
Currently collecting data from YouTube and Facebook
Miriam Marciel, ESR3 NEC Labs
Valuating the users

Our preliminary results in YouTube show:

Prices vary per country
Miriam Marciel, ESR3 NEC Labs
Valuating the users

Prices vary per category
Miriam Marciel, ESR3 NEC Labs
Valuating the users

Prices vary per gender
Miriam Marciel, ESR3 NEC Labs
Valuating the users

Prices vary per age
Miriam Marciel, ESR3 NEC Labs
Future research

Value of the user
 Variation of prices

Developing tools to predict prices for advertisers

Provide users quantification in revenue of their online
activity
Miriam Marciel, ESR3 NEC Labs
Training

PhD program: UC3M

Online courses
- Intro to Data Science (Udacity)
- Intro to Machine Learning (Udacity)
- Machine Learning (Coursera)
- Mining Massive Datasets (Coursera)

Attendance to conferences:
- NSDI'15
- WWW'15
Miriam Marciel, ESR3 NEC Labs
Training inside NEC

Industrial training:
- Information Security Management System
- Business model communication

Technical training:
- Colloquium talks
Miriam Marciel, ESR3 NEC Labs
Career development

Continue working in an R&D environment
 Developing services for companies and users
Miriam Marciel, ESR3 NEC Labs
STUDENT PRESENTATIONS
• Roberto Gonzalez, ER1 NEC Labs
• Supervisors: Saverio Niccolini/Mohamed Ahmed
Roberto Gonzalez, ER1 NEC Labs
Presentation


Msc. In Telecommunitacion Engineering from Universidad
Carlos III de Madrid in 2009
Msc. In Telematic Engineering from Universidad Carlos III de
Madrid and Polytechnic University of Cataluña 2011

Phd. In Telematic Engineering from Universidad Carlos III de
Madrid 2015

6 months internship at the University of Oregon in 2013
Roberto Gonzalez, ER1 NEC Labs
My work in the project

WP2: Big Data Analysis for Advanced Network Management
& Operation

Task 2.1: Performance optimizations of distributed stream
processing platforms

Task 2.2: Stream mining of content popularity lifetime
evolution and its applications

Task 2.3: Cross-device privacy preserving user
identification

Supervisors:

NEC: Saverio Niccolini/Mohamed Ahmed
Roberto Gonzalez, ER1 NEC Labs
How to make it


Analysis of real network data to understand the advertising
ecosystem

Protect the privacy of the users

Streaming ad matching

Users and content popularity/price
Analysis of social networks data
Roberto Gonzalez, ER1 NEC Labs
Research activities and results

Measurement and analysis of the online advertising ecosystem:

Analysis of the cookies in the wild (Under review at CCR Jan'16
edition)

Online advertising matching (Under review at PKDD journal 12/15)

Measuring fraud in Video Content Portals (submission being
prepared)


Measuring value of the user on the Internet (current work)
Analysis of social networks (1 paper published in TON and 1 in PLOS
ONE)
Roberto Gonzalez, ER1 NEC Labs
Work overview

The advertising market is growing and increasing in complexity:

Ad networks, DSPs, SSPs, Data suppliers, re-targeters…



We don’t understand who trades user data and when
User data my be resold many times, move jurisdictions and etc.
User don’t really understand
http://www.displayadtech.com/the_display_advertising_technology_landscape
Roberto Gonzalez, ER1 NEC Labs
Work overview
Roberto Gonzalez, ER1 NEC Labs
Work overview
Roberto Gonzalez, ER1 NEC Labs
Work overview


We don't know how the users are tracked

Analysis of the cookies

Cookies are complex and can be used in different ways
Let's start analyzing how the cookies are used

One month of network data of about 13K users.

2.2B HTTP requests → 3 Tb of data!!
Roberto Gonzalez, ER1 NEC Labs
Main findings

Tracking cookies reach most of the users
Roberto Gonzalez, ER1 NEC Labs
Main findings

There are domains setting cookies in behalf of others (false first
party cookies)

More than 60K domains set Google Analytics cookies

Some advertisers serve ads as a subdomain to avoid being
blocked
Roberto Gonzalez, ER1 NEC Labs
Main findings

Each cookie can use a different format

We know how to analyze more than 85% of them!!
Roberto Gonzalez, ER1 NEC Labs
Future research


Automatic detection of tracking on the Internet using network
data
Design of solutions able to modify the network data in a
streaming way to protect the users
Roberto Gonzalez, ER1 NEC Labs
Training


Industrial training:

Information Security Management System

Business model communication

Exposition to Business proposals
Technical training:


Colloquium talks
Attendance to conferences:

NSDI'15

WWW'15
Roberto Gonzalez, ER1 NEC Labs
Career development

Data analytics and research

Understand the advertising ecosystem

Build solutions to help Internet users protecting their online
privacy.

Actively participate in the design of new business opportunities
for NEC
Roberto Gonzalez, ER1 NEC Labs
STUDENT PRESENTATIONS
Maria Ines Robles, ESR8 Ericsson
Supervisor: Petri Jokela, Jörg Ott
Background

Graduated in Universidad Tecnologica Nacional – Facultad Regional Mendoza
(UTN-FRM) – Argentina

Intership for 6 months in Frankfurt University of Applied Sciences - Germany

Worked in the Industry 3 years after graduation.

Teaching at University – (UTN -FRM)

Researcher at UTN - GridTICS Lab (Grupo de Investigación y Desarrollos en
TICs).

Co-chair in IETF-ROLL (Routing over Low Power and Lossy Networks)
Working group.

PhD Student at Aalto

Working in Ericsson.
Maria Ines Robles, ESR8 Ericsson
Research work

PhD METRICS Position 8 – Ericsson LMF (Finland)

Understanding Machine-to-Machine (M2M) communication performance

Supervisor at Aalto: Prof. Dr. Jörg Ott

Supervisor at Ericsson: Petri Jokela
Maria Ines Robles, ESR8 Ericsson
Training
- Internal Ericsson Training in M2M technologies:
- Bluetooth Low Energy,
- 802.11ah
- Lightweight M2M (LWM2M)
- IPSO (IP for Smart Object).
- Ethical Research course at Aalto University
Maria Ines Robles, ESR8 Ericsson
Training
-Present Activities:
- Member of IETF – Internet-of-Things (IoT) Directorate
- Continue with co-chair of IETF ROLL WG.
Past Activities during METRICS project:
- Member of Technical Program Committee (TPC) for EAI International Conference on Interoperability in IoT –
October 2015.
- Co-chair of the Low Power and Lossy Networks Plugfest at IETF90 - Report was published in IETF Journal
- Training to the Routing IETF Chairs based on RFC 4858 - Document Shepherding from Working Group
- Document Shepherd of the following IETF I-D:
- Forwarder policy for multicast with admin-local scope in the Multicast Protocol for Low power and Lossy
Networks (MPL)
- MPL Parameter Configuration Option for DHCPv6
- IETF Anti-Harassment Procedures
Maria Ines Robles, ESR8 Ericsson
M2M IoT Novel Meassurements and
Evaluation Methods- WP1
Context to apply the Measurements → Application Layer of M2M:
- Protocols to transfer data between constrained devices: Constrained Application Protocols (CoAP)
- Protocol to manage constrained devices: Lightweight M2M (LWM2M):
- Entities: Client, Server and Proxy (Functions of Server and Client)
- Data Model providing semantic meaning: IPSO Objects (IP for Smart Object)
Maria Ines Robles, ESR8 Ericsson
M2M IoT Novel Meassurements and
Evaluation Methods- WP1
Context to apply the Measurements → Application Layer of M2M:
- Protocols to transfer data between constrained devices: Constrained Application Protocols (CoAP)
- Protocol to manage constrained devices: Lightweight M2M (LWM2M):
- Entities: Client, Server and Proxy (Functions of Server and Client)
- Data Model providing semantic meaning: IPSO Objects (IP for Smart Object)
- Analyzing:
- IP Compression for Low Layer (802.11ah)
- Use Cases of Routing for M2M
Maria Ines Robles, ESR8 Ericsson
M2M IoT Novel Meassurements and
Evaluation Methods- WP1

Currently analyzing the scalability of the LWM2M:
–

Proxy and without Proxy:
•
Effect on power consumption
•
Effect on traffic
Hardware: Raspberry Pi and Intel Edison.
Maria Ines Robles, ESR8 Ericsson
Publications
- Design of a Performance Measurements Platform in Lightweight M2M for Internet of
Things
- To be presented at Workshop on Research and Applications of Internet Measurements (RAIM)
in Cooperation with ACM SIGCOMM – November 2015
- IPv6 over 802.11ah
- Presented at IETF 93
- Presented also in IEEE 802.11 Wireless Next Generation (WNG) 2015 Meeting
- New version to be presented at IETF 94
- When to use RFC 6553, 6554 and IPv6-in-IPv6
- Presented at IETF 93.
- New version to be presented in IETF 94.
- Low-power, Lossy Network Plugfest Demonstrates Running Internet of Things Code
- IETF Journal - November 2014 - Volume 10, Issue 2- Pag., 18 -20.
Maria Ines Robles, ESR8 Ericsson
Future Steps

Install a test bed to determinate traffic patterns with LWM2M.

Evaluate compression efficiency for 6LoWPAN over 802.11

Analyze the behavior of related protocols

Continue with IETF work: ROLL, 6Lo, t2trg.
Maria Ines Robles, ESR8 Ericsson
Future Career
Focus in Research on IoT topics
Protocols Design: Participation in Standardization bodies
Maria Ines Robles, ESR8 Ericsson
STUDENT PRESENTATIONS
Roberto Morabito, ESR9 Ericsson
Supervisor: Jörg Ott, Nicklas Beijar
Personal Background
2010 – Bachelor of Science (B.Sc.) in Telecommunication Engineering
at Università Mediterranea di Reggio Calabria (Italy)
Thesis Title: “Small Scale Propagation Models for DVB-T Systems”
2008 – Visiting Student at Universitat Politècnica de Catalunya
Escola Tècnica Superior d'Enginyeria de Telecomunicació de Barcelona (ETSETB)
2013 – Master of Science (M.Sc.) in Computer and Telecommunication
Systems Engineering at Università Mediterranea di Reggio Calabria (Italy)
Thesis Title: “OpenFlow Impact on Wireless Networks”
2013 – Visiting Student at Brunel University of London
School of Engineering and Design - Wireless Network Communications Centre
Roberto Morabito, ESR9 Ericsson
Roberto Morabito (ESR 9) – Ericsson
D1.1
• Novel Measurement and
Evaluation Methods
D1.2
• Measurements using
new methods
D1.3
• Evaluation of Results
D2.1
D2.2
D2.3
• Identification of Large Dataset and
requirements for processing
• New tools for large dataset
processing
• Application of new tools to large
data set
WP1
WP2
Supervisor: Jörg Ott
Thesis Advisor: Nicklas Beijar
Roberto Morabito, ESR9 Ericsson
Understanding and Characterizing Distributed
Cloud Performance in Capillary Networks
› Data Centers can play an important role for off-line
analysis of large amount of data
› Different actions are required for real-time operations
› Network performance can become the bottleneck for
particular IoT services
Roberto Morabito, ESR9 Ericsson
Understanding and Characterizing Distributed
Cloud Performance in Capillary Networks
› Distributed Cloud topology
can meet the
requirements for our
scenario
› Cloud services are not
only able to run in a data
center but also close to
the device itself (IoT
Gateways)
Roberto Morabito, ESR9 Ericsson
Understanding and Characterizing Distributed
Cloud Performance in Capillary Networks
› Some of the operations made at the gateway layer have
very strict requirements
› We need Lightweight (and versatile) solutions to match
the scenario requirements
Roberto Morabito, ESR9 Ericsson
Lightweight Virtualization
LXC
Minimal
Performance
Overhead
Fast Boot and
Shutdown
Smaller File
Sizes
Roberto Morabito, ESR9 Ericsson
Can run a
“Light” Cloud
Understanding and Characterizing Distributed
Cloud Performance in Capillary Networks
1
• Identification of measurements
challenges in the analyzed scenario
2
• Definition and categorization of
relevant performance indicators
3
• Definition and categorization of
relevant evaluation methods
Roberto Morabito, ESR9 Ericsson
Understanding and Characterizing Distributed
Cloud Performance in Capillary Networks
METRICS GROUP
Performance Indicator
Processing
Runtime
MIPS/mW
MFlops/mW
MFlops
Storage
Storage
IOPS (Input Output Operations Per
Second). Random seek
Memory
Memory Throughput
Network
Network throughput
Packet loss
Power Consumption
Power Consumption
Power efficiency
Energy Efficiency
Roberto Morabito, ESR9 Ericsson
Understanding and Characterizing Distributed
Cloud Performance in Capillary Networks
Roberto Morabito, ESR9 Ericsson
Understanding and Characterizing Distributed
Cloud Performance in Capillary Networks
Roberto Morabito, ESR9 Ericsson
Obtained Results
Roberto Morabito, ESR9 Ericsson
Obtained Results
Roberto Morabito, ESR9 Ericsson
Obtained results
Roberto Morabito, ESR9 Ericsson
Obtained results
Roberto Morabito, ESR9 Ericsson
Obtained results
Roberto Morabito, ESR9 Ericsson
Obtained results
Roberto Morabito, ESR9 Ericsson
Ongoing work
• Profile the Power Consumption of Virtualization
Technologies, when running on embedded systems
(Raspberry Pi 2/Intel Edison, etc.)
Roberto Morabito, ESR9 Ericsson
Future Plans of the Research Project
Short/Mid term view
Performance comparison, to be
performed at the capillary
gateway layer, of several data
compression/filtering/aggregation
methods.
Performance estimation of
different orchestration tools built
around the virtualization
technologies.
Roberto Morabito, ESR9 Ericsson
Include the usage of network
simulators in our research, to
evaluate all the benefits and the
possible trade-offs introduced by
distributed cloud in IoT contexts
on a larger scale.
Future Plans of the Research Project
Long term view
Replacement of manual management
processes with automated ones,
capable of distributed management of
massive amounts of data
Deployment and execution of virtualized
processing functions in suitable places
in a distributed cloud network, based on
resource availability and other
constraints, policies, optimization
Roberto Morabito, ESR9 Ericsson
Obtained results
Publications
Roberto Morabito, Jimmy Kjällman, and Miika Komu.
“Hypervisors vs. Lightweight Virtualization: a Performance Comparison”
2015 IEEE International Conference on Cloud Engineering (IC2E 2015), Tempe
(Arizona).
Roberto Morabito
“Power Consumption of Virtualization Technologies: an Empirical Investigation”
Submitted to IEEE/ACM International Conference on Utility and Cloud Computing (UCC
2016)
Standard Contributions
IETF Draft: IPv6 over 802.11ah – draft-delcarpio-6lo-wlanah-00
•
•
Presented at the IETF 93 Meeting (Prague, July 2015)
Presented at the IEEE 802.11 Wireless Next Generation (WNG) 2015
Meeting
Roberto Morabito, ESR9 Ericsson
Training
• Enrolled as PhD student at Aalto University (Finland)
• Ericsson Learnathon: in Ericsson Finland (LMF) is usually a quarterly
event. Can be defined as a learning hackathon/marathon. The main idea is
to share knowledge and transferring skills between employees.
• Ericsson Cloud Learnathon: It has the same purpose of the abovementioned Learnathon, but in this case the learning objectives are
increasing knowledge, awareness, and understanding of Cloud Computing
in several aspects.
• Ericsson Internal Tech Talk: “Network Performance of Virtualization
Technologies”.
• Ericsson Internal Tech Talk: “Linux Containers: The current status and
future directions”.
• Ericsson Internal Tech Talk: “Cloud Simulations using CloudSim”.
• Member of Docker Helsinki Meetup
Roberto Morabito, ESR9 Ericsson
Training
•
•
•
•
•
•
•
•
E-learning Course: “Software Defined Networking”– Coursera
E-learning Course: ”R programming”
E-learning Course: “Cloud Computing Concepts (part I)”
E-learning Course: “Fog Networks and the Internet of Things”
E-learning Course: “Cloud Computing Concepts (part II)”
International Workshop on Container Technologies and Container Cloud
International Conference on Cloud Engineering (IEEE IC2E)
Technical Program Committee for 2015 IEEE Global Communications
Conference (GLOBECOM) in the Software Defined Networking and
Network Function Virtualization track.
Roberto Morabito, ESR9 Ericsson
Future Career Plans
Working in Research (Industry or Academia)
Roberto Morabito, ESR9 Ericsson
STUDENT PRESENTATIONS
Costas Iordanou, ESR6 Telefonica R&D
Supervisor: Nikolaos Laoutaris, Ruben Cuevas
About my self
 Cyprus University of Technology
• (2009 -2013) Bachelor's degree, Computer Engineering and Informatics
Publications:
1. Hermes: Architecting a Top-Performing Fault-Tolerant Routing Algorithm for
Networks-on-Chips, Networks-on-Chip (NoCS), 2014 Eighth IEEE/ACM International
Symposium on , vol., no., pp.178,179, 17-19
2. Hermes: Architecting a Top-Performing Fault-Tolerant Routing Algorithm for
Networks-on-Chips, Computer Design (ICCD), 2014 32nd IEEE International
Conference on , vol., no., pp.424,431, 19-22
Awards:
o Second best score award.
• (2013-2014) MSc in Computer Engineering and Informatics
Awards:
o Full scholarship
o Financial support from the department of computer engineering and informatics.
 Universidad Carlos III de Madrid
• (2014-2015) Master’s Degree, in Telematic Engineering
Costas Iordanou, ESR6 Telefonica R&D
Research Topic – WP2 D2.2 M2.2
Online Privacy and
Targeted Advertisements
Costas Iordanou, ESR6 Telefonica R&D
Research Goals and Progress
• 3 Main Goals:
• Measurements on web tracking and personalized services on the web.
• Establishing causality between observed targeted advertisements and past
browsing behavior.
• Developing tools and obfuscation techniques for preserving one’s privacy
in view of revealed targeted advertisement methods.
• Research progress (2 Tasks):
1. Profile based price discrimination detection tool - $heriff (
http://sheriff.dynu.com ).
2. Web Identity Translator – Behavioral advertising and identity privacy with
WIT. (Under review - HotNets2015, Co-author).
Costas Iordanou, ESR6 Telefonica R&D
Task 1 – $heriff Tool
Research question:
Detect if profile based price discrimination is tacking place on the web, and if yes,
by whom, and how?
Current State-of-the-Art:
1. Mikians et. al. – Detect price discrimination based on user location.
2. Hannak et. al. – Detect also profile based price discrimination in a small scale
Our approach:
• Large scale measurements by
• allowing all Internet users to check product prices from
• any available web store.
https://addons.mozilla.org/en-US/firefox/addon/sheriff_v2/
Costas Iordanou, ESR6 Telefonica R&D
Task 1 – $heriff Tool (Promising results)
User profile 1
User profile 2
Costas Iordanou, ESR6 Telefonica R&D
Task 2 – Web Identity Translator
Research questions:
(Under review - HotNets2015, Co-author)
1. To what degree does browsing history uniquely identify users?
2. To what extent can we intervene to reduce users’ identifiability?
3. Does this intervention allow advertisers to infer users interests?
Our approach:
Client Browser
Private cookies
Results:
WIT Proxy
Private cookies
Web Site
Public cookies
Tracker cookie
Requests Intervention
Identifiable users protection
10 %
70 %
20 %
90 %
With minimal impact on getting relevant ads after
intervention based on behavioral tags analysis.
Costas Iordanou, ESR6 Telefonica R&D
User Mapping
Training and additional actions
• Training
• MSc in Telematic Engineering (UC3M)
• Weekly meetings for the Data Transparency Lab
(http://datatransparencylab.org/). My contribution – Developing tools towards
data transparency.
• Mobile World Congress 2015, Barcelona, Spain – Tool Demo.
• Telefonica learning week. Technical presentations, tools and methodologies.
• Seminar from Bruce Maggs (Akamai) at UC3M.
• Conferences:
• DTL conference 2014, Barcelona, Spain
• Sigcomm 2015, London, United Kingdom
• Collaboration with journalists
• Julia Angwin, senior reporter – http://www.propublica.org
• Anne Debroise, free-lance journalist – http://www.science-et-vie.com/
• Kate Abnett, editorial Associate – http://www.businessoffashion.com/
• Cathy O'Neil, author of the blog – http://mathbabe.org/
Costas Iordanou, ESR6 Telefonica R&D
Hard Technical Points
Measurements needed to be taken at the user side (web
browser):
• Overcome HTTPS traffic limitations (Google, Facebook, Etc.)
• Detect user side tracking techniques (JavaScript cookies, fingerprinting code,
etc.)
Solution:
• Specialization on web browsers architecture and APIs.
• Develop browser plugins for Mozilla Firefox and Google Chrome.
• Monitor common tracking techniques and PII leakage from the user side.
• Overcome HTTPS limitations.
• Accurate measurements.
• User consent.
• Etc.
Costas Iordanou, ESR6 Telefonica R&D
Future Work and Career
Future work:
• $heriff tool – Get more users, collect data and publish my work.
• Online Behavior Advertisement (OBA) reverse engineering tool. Starting this
month.
Future Career:
Work with regulators: Monitoring organizations and web services on how they
handle personal data and privacy issues.
Academic career on web technologies to protect user privacy on the web.
o Alternative ways to provide privacy guarantees without compromise user
experience (break web sites).
Costas Iordanou, ESR6 Telefonica R&D
STUDENT PRESENTATIONS
Leonardo Tonetto - ESR 7 TID
Supervisor: Alessandro Finamore
(My) Background
1987.
Born in Itumbiara-GO, Brazil
2005-2010.
Bsc. Electrical Engineering @ UFU, Brazil
2010-2012.
Software Engineer
2012-2014.
Msc. Distributed Systems @ UPC+KTH, Spain+Sweden
2015-Now.
ITN METRICS, ESR 7 @ TID, Barcelona - Spain
LeonardoTonetto
Tonetto - ESR 7 TID
@ TID + TUM
Leonardo
90
Research Work - Motivation
• MOBILITY
LeonardoTonetto
Tonetto - ESR 7 TID
@ TID + TUM
Leonardo
91
Research Work
• Cellular Network Data
–
Mobility information
–
Web logs
• Algorithms to extract Mobility features
• Correlation with other sources of Data
– Demographics
– Client information
LeonardoTonetto
Tonetto - ESR 7 TID
@ TID + TUM
Leonardo
92
(Preliminary) Results
Leonardo
7 TID
LeonardoTonetto
Tonetto -- ESR
ESR 7
@ TID + TUM
93
Future Work
• Correlate Mobility with User Experience
• Framework to optimize future Network
Deployments
LeonardoTonetto
Tonetto - ESR 7 TID
@ TID + TUM
Leonardo
94
Challenges
• Data characteristics
–BIG Data -> Better algorithms
–Data Quality -> Data Cleaning
LeonardoTonetto
Tonetto - ESR 7 TID
@ TID + TUM
Leonardo
95
Accomplishments
1 Achievements
• Work in progress
planned to be
published at PAM
2015
• Deep understanding
of the problem and
the data set used
2 Training
• Online self-paced
course in Machine
Learning
• Attended technical
talks @ TID given by
co-workers and
guests
Attended
Conferences
3
• TMA 2015,
Barcelona
• SIGCOMM 2015,
London
• New programming
languages learned
LeonardoTonetto
Tonetto - ESR 7 TID
@ TID + TUM
Leonardo
96
Career Prospects
• Continue as a researcher
–Universities, R&D Departments
• Industry
–Systems Engineer
• Regulatory Agencies
LeonardoTonetto
Tonetto - ESR 7 TID
@ TID + TUM
Leonardo
97
STUDENT PRESENTATIONS
Alemnew Asrese, ESR1 AALTO
Supervisor: Jörg Ott
Who am I?
• (Now) Doctoral Candidate at Aalto
• (2014) MSc. in Computer Science – UniTN - IT
• (2010) BSc. in Information Science – Adama Sc. &
Tech. University - ET
• Intern at NEC Labs
• Web developer & programmer at Green Prefab Italia
• Graduate Assistant at ASTU
Alemnew Asrese, ESR1 AALTO
Understanding and Improving
Mobile User Experience
WP1: Advanced Instrumentation for
Future Generation IP Networks
– Measurement and evaluation Methods
and data collection tools
– Carry out measurement
– Evaluate results
Alemnew Asrese, ESR1 AALTO
Quality of Experience
“QoE is the degree of delight or annoyance of the user of the
application or the system...” -- Qualinet white paper
– Subjective assessment
• Asking a user how he feels about the service &
rate it with MOS
• Not scalable and resource consuming
– Objective measures
• Using technical parameters such as PLT for
websites
• Hard to get the real user feeling
Alemnew Asrese, ESR1 AALTO
The Web QoE
• The web becomes more complex
– Technology advancement
• More JS, Style Sheets, etc
– Increase in the user requirement
• The user want the website to be dynamic
• This complexity affects the user experience of the
web application
Alemnew Asrese, ESR1 AALTO
What affects the Web QoE?
•
•
•
•
•
Address resolution
Internet transport
Content placement
Content caching
Content structure
Alemnew Asrese, ESR1 AALTO
A measurement tool
to infer Web QoE
• WebPerf: measures the technical parameters
– DNS lookup time, Number of DNS iterations, TCP connect
time
– TLS hand shake time, latency, download time, number of
HTTP elements & redirects, …
– Network PLT
Alemnew Asrese, ESR1 AALTO
A measurement tool
to infer Web QoE
• Web Rendering:
– Measures the rendering time of a website
– Based on the pixel changes at the browser window
– ~ Perceived PLT
Alemnew Asrese, ESR1 AALTO
The Web Performance and rendering time
measurement tool has a distributed architecture
Alemnew Asrese, ESR1 AALTO
Rendering time of websites
Alemnew Asrese, ESR1 AALTO
Current status and results
• The tool has been deployed and collecting data
• A paper about the tool and the results is almost ready
for submission (target: PAM 2016)
• A poster has presented at TMA 2015 PhD school,
Barcelona
Alemnew Asrese, ESR1 AALTO
Open Issues
• How to precisely define when a page has fully
rendered
– Use machine learning techniques
• How to correlate these objective measures with the
users’ feeling
– Conduct a user study in a controlled environment
Alemnew Asrese, ESR1 AALTO
Trainings
• Dagstuhl Seminar on “QoE: from Assesment to
Application”
– Output: a paper to QoMEX 2015
– Conferences: IMC 2014, PAM 2015, TMA 2015, WWW 2015,
SIGCOMM 2015, IMC 2015 (travel grant)
– METRICS events
– Doctoral Courses in Aalto:
•
•
•
•
•
TA for Communication Networks
TA for Network Programming
Quality of Experience
Computer Networks: Advanced features
Internet Traffic Measurement and Analysis
Alemnew Asrese, ESR1 AALTO
What is next?
• Extend the tool for mobile platforms and
collect data from mobile users
• QoE in Web RTC and video streaming
services
• Study the factors that affect QoE in mobile
networks
Alemnew Asrese, ESR1 AALTO
My future?
• Still under investigation (~3 years yet )
• I most probably stay in the research and academia
environment
Alemnew Asrese, ESR1 AALTO
STUDENT PRESENTATIONS
• Ermias Walelgne, ESR 2 AALTO
• Supervisor: Jörg Ott, Jukka Manner
Background
•Education:
BSc in Computer Science: Hawassa University, Ethiopia - 2008
MSc in Computer Science: University of Trento, Italy - 2014
(Data management and internet technologies)
Work: Graduate Assistant: University of Gondar, Ethiopia - 2008 to 2010
Assistant lecturer Bahir Dar University, Ethiopia - 2010 to 2011
Intern: Junior Researcher, at University of Illinois Research Center
(Advanced Science Research Centre): Singapore - August-Nov. 2012
Ermias Walelgne, ESR 2 AALTO
Metrics
•WP1: Advanced Instrumentation for Future
Generation IP Networks
Novel measurement and evaluation methods
Develop and extend tools to support data collection
• ESR2:
– Characterizing wireless network performance
– Modelling elements of user behaviour following up from the traces.
• Supervisors: Joerg Ott , Jukka Manner
Ermias Walelgne, ESR 2 AALTO
Netradar
•Free mobile network measurement service using crowdsourcing
•Developed at Aalto University since 2010
•Measurement points distributed globally (servers in EU, US and Asia)
•Measure and collect information:
•
•
•
•
•
Signal quality and strength per radio technology
Download/upload speed (sustainable and peak TCP)
Detailed TCP metrics and Latency (UDP)
Location and Time stamps
Base station information and Cellular radio technology
Operator name, country (MNC, MCC) and Device information (hardware, OS)
Ermias Walelgne, ESR 2 AALTO
Factors for cellular network performance
•Dependence on user mobility
Time variant behavior
•
•Access technology and the device
•
•
Impacts of user's own action
Ermias Walelgne, ESR 2 AALTO
Characterizing cellular network performance
•
•
•
Study different factors for received bandwidth quality:
[radio technology, network coverage, congestion caused by user's,
mobile device itself and handovers]
•
Performance of various service providers, operators, mobile devices ...
•
Obtain user’s experience based on mobility, time and location
•
Study the effect of handover and stability of the network
•
Ermias Walelgne, ESR 2 AALTO
Motivation of problem
Bit rate suddenly dropped to zero
and stay dropped for a while
Ermias Walelgne, ESR 2 AALTO
Motivation of problem
•1784602: always stationary - Cell type network measurements
•538639: suddenly drop to zero bit rate and lasted for at
least 200ms in consecutive
• 30.18% of the measurement
•Why this is a problem?
•
Depending on the application the user is running this
could cause service interruption [ eg. Online games, Video streaming ]
Ermias Walelgne, ESR 2 AALTO
Some statistics
•Among total sudden drop stationary cellular type
networks
•
31.30% change their radio technology [Vertical handovers]
•
27.96% change their cell ID [Horizontal handover]
Vertical handovers are more frequent than horizontal
Poster presented: TMA 2015 PhD school, Barcelona
Ermias Walelgne, ESR 2 AALTO
On such sudden drop vertical
handover provide less latency
Fellow Number: ESR2 | Institution: Aalto | Supervisor: Prof. Dr.-Ing. Joerg Ott
Ermias Walelgne, ESR 2 AALTO
Bit rate sudden drop average duration in Helsinki
Ermias Walelgne, ESR 2 AALTO
Ermias Walelgne, ESR 2 AALTO
Training activities
•Conferences attended:
PAM 2015, TMA 2015, SIGCOMM 2015
Courses:
Network Traffic measurement and analysis,Network Services
Business Models,
QoE reading group meetings
TA: Protocol Design
Training and Tutorials:
TMA Phd summer school, Cloud Storage, Open Hardware Networking
Workshop:
Ethics in Networked Systems Research: ACM SigComm
Ermias Walelgne, ESR 2 AALTO
Future plans:
Extending the measurement tool
More detailed metrics e.g. Traceroute
User experience modeling
Consumption profile of user activity
Service based QoE: [ given a certain metrics values eg. RTT and
bandwidth what is my expected quality of experience for this service ]
Modelling elements of user behaviour
Ermias Walelgne, ESR 2 AALTO
Future Carrier
Continue working on internet measurement and mobile
data analysis
•
•
Academic and R&D centers
Ermias Walelgne, ESR 2 AALTO
STUDENT PRESENTATIONS
• Juan Brenes, ESR 12 UC3M
• Supervisor: Marcelo Bagnulo
Juan Brenes, ESR 12 UC3M
Personal Background
Telematic Engineer, Universidad de Montevideo,
Montevideo, Uruguay 2005-2011

Assistant Professor, Universidad de Montevideo,
Montevideo, Uruguay 2008-2013

Network Administrator & Developer, Canal 10,
Montevideo, Uruguay 2010-2014

Juan Brenes, ESR 12 UC3M
Metrics

ESR Number: 12

Project: Data path vs control path measurements

Institution: UC3M

WP1: Novel measurement and evaluation methods

Supervisor: Marcelo Bagnulo
Juan Brenes, ESR 12 UC3M
Data path vs control path
measurements

Gather Data/Control plane metrics

Correlate metrics

Develop tools for network monitoring
Juan Brenes, ESR 12 UC3M
First step - Data path? Vs control
path? measurements



AIM THE TARGET!
Data and control planes
are changing (NFV, SDN,
SFC)
Working on NFV
Juan Brenes, ESR 12 UC3M
Identify traffic characteristics

Traffic from and to
external networks

Packets bouncing from
server to server
Juan Brenes, ESR 12 UC3M
Second step – Develop the tests!
- Use state of the art
topologies (Fat-tree,
Bcube and Jellyfish)
- Simulate NFV Data
plane traffic
Juan Brenes, ESR 12 UC3M
Measurements?
•UDP as transport protocol
•Bounce packets between servers
•Change the network load
•Measure: Packets traversing,
delay and jitter
Juan Brenes, ESR 12 UC3M
Third Step – Gather the data!
PACKETS TRAVERSING
Juan Brenes, ESR 12 UC3M
Packet Delay (ns)
Number of traffic generators
Juan Brenes, ESR 12 UC3M
Jitter (ns)
Number of traffic generators
Juan Brenes, ESR 12 UC3M
Mathematical analysis


Amount of devices
Shortest paths
Edge disjoint paths

“Reachability”

Juan Brenes, ESR 12 UC3M
Jellyfish – Reachability
Juan Brenes, ESR 12 UC3M
Topologies do not perform properly
under NFV traffic

Jellyfish performs better than Fat-tree and Bcube.

Jellyfish uses less network resources.
Juan Brenes, ESR 12 UC3M
Future steps – Answer some missing
questions

Why and where does the congestion
happen?

Can topologies be improved for NFV?
Juan Brenes, ESR 12 UC3M
Training





Broadband networks UC3M – Telematics Master degree
Advanced Statistical Inference UC3M – Mathematics
Master degree
Optimization UC3M – Mathematics Master degree
Simulation of networks and Telematics applications UC3M
– Telematics master degree
Tools for research in Telematics Engineering UC3M –
Telematics master degree
Juan Brenes, ESR 12 UC3M
Training






Network complexity analysis UC3M – Telematics master
degree
Mobile Networks UC3M – Telematics master degree
Architecture of the Internet UC3M – Telematics master
degree
Content Delivery Networks UC3M – Telematics master
degree
Internet Measurement UC3M – Telematics master degree
Mobility, Security and Usability Aspects in Ubiquitous
Computing and the Future Internet UC3M – Telematics
master degree
Juan Brenes, ESR 12 UC3M
STUDENT PRESENTATIONS
• Anna Maria Mandalari, ESR 13 UC3M
• Supervisor: Marcelo Bagnulo
Anna Maria Mandalari, ESR 13 UC3M
METRICS ITN:WP1 Objectives
•
Provide novel measurements and evaluation
methods
•
Carry out measurements using existing (own)
measurement infrastructure
•
Evaluate the results
Anna Maria Mandalari, ESR 13 UC3M
METRICS ITN:WP1 Objectives
•
Provide novel measurements and evaluation
methods
•
Carry out measurements using existing (own)
measurement infrastructure
•
Evaluate the results
Anna Maria Mandalari, ESR 13 UC3M
Identifying network performance
parameters that reflect QoE
Goal of my research: New methodologies for measuring
and improving the QoE
Is the Internet
Ossified?
How does it affect the QoE?
Anna Maria Mandalari, ESR 13 UC3M
Is the Internet Ossified?
Anna Maria Mandalari, ESR 13 UC3M
Identifying network performance
parameters that reflect QoE
Criticism:
Middleboxes behavior
Anna Maria Mandalari, ESR 13 UC3M
Identifying network performance
parameters that reflect QoE
• How will Internet react to a new protocol?
Understand the interaction of the new
solutions with the middleboxes active
along the path.
Anna Maria Mandalari, ESR 13 UC3M
METRICS ITN:WP1 Objectives
•
Provide novel measurements and evaluation
methods
•
Carry out measurements using existing (own)
measurement infrastructure
•
Evaluate the results
Anna Maria Mandalari, ESR 13 UC3M
How to measure a thousand
end-users?
•
•
•
Be Google (or any other large Internet players)
Use existing large-scale measurements platforms
Get your code to run on a thousand users’ machines through
another delivery channel
Anna Maria Mandalari, ESR 13 UC3M
Existing Large-Scale
measurements platforms
Anna Maria Mandalari, ESR 13 UC3M
Existing Large-Scale measurements
platforms: limitations
•
The limited and often special position of
testbed nodes
•
No possibility to deploy your own test
•
Fixed line only
•
Access to the results
Anna Maria Mandalari, ESR 13 UC3M
Crowdsourcing platform
Perform large-scale Internet measurement campaigns
Anna Maria Mandalari, ESR 13 UC3M
Crowdsourcing platform:
Advantages
•
•
Possibility to select dedicated participants basing on
certain criteria: countries, devices etc..
Very high diversity of the users (geographical locations
and multiple ISP networks)
•
Possibility to deploy your own test
•
Access to mobile network
•
Low cost measurements
•
Controllability of the probes
Anna Maria Mandalari, ESR 13 UC3M
Crowdsourcing platform:
Advantages
• Large-scale and realistic QoE measurements
Anna Maria Mandalari, ESR 13 UC3M
Crowdsourcing platform:
Challenges
•Diversity
•Trusted
of end-user devices
Clients
Anna Maria Mandalari, ESR 13 UC3M
METRICS ITN:WP1 Objectives
•
Provide novel measurements and evaluation
methods
•
Carry out measurements using existing (own)
measurement infrastructure
•
Evaluate the results
Anna Maria Mandalari, ESR 13 UC3M
QoE: Case Studies
•
The case of the pervasive encryption
•
TCP Fast Open
•
HTTP/2
Anna Maria Mandalari, ESR 13 UC3M
QoE: Case Studies
•
The case of the pervasive encryption
•
TCP Fast Open
•
HTTP/2
Anna Maria Mandalari, ESR 13 UC3M
Case Study: The case of Pervasive
Encryption
Challenge: Provide encryption by default for all Internet communications
Naylor, David, et al. "The Cost of the S in HTTPS." Proceedings of the 10th ACM International on Conference on emerging Networking
Experiments and Technologies. ACM, 2014.
Anna Maria Mandalari, ESR 13 UC3M
Results
ERROR = (success [HTTP] − success [TLS]) / success [HTTP]
70% of the users that use a proxy are not able to perform a TLS connection over port 80 in mobile network.
164
Anna Maria
Mandalari, ESR 13 UC3M
Case Study: The case of
Pervasive Encryption
•
25% of the users are not able to perform a TLS connection over port
80 in mobile network;
•
70% of the users that use a proxy are not able to perform a TLS
connection over port 80 in mobile network;
•
It is probably feasible to roll out TLS protection for most ports except
for port 80, assuming a low failure rate (6%);
Anna Maria Mandalari, ESR 13 UC3M
Case Study: The case of
Pervasive Encryption
These results can serve as a lower bound for the
failure rate for using protocols other than expected in
different ports.
Anna Maria Mandalari, ESR 13 UC3M
Publications
•
A. M. Mandalari, M. Bagnulo, A. Lutu. "Informing Protocol Design Through
Crowdsourcing: the Case of Pervasive Encryption". ACM SIGCOMM
Workshop on Crowdsourcing and crowdsharing of Big (Internet) Data (C2B(I)D),
Aug 2015, London. ACM SIGCOMM Computer Communication Review, 2015.
[Best paper award at C2B(I)D 2015].
•
Special issue of SIGCOMM CCR that will include the proceedings of SIGCOMM
2015 and award papers from other workshops
•
It has been accepted to be presented at JNIC2015 in Leon, Spain
•
It has been accepted to be presented at RAIM 2015 Workshop in cooperation
with ACM SIGCOMM in Yokohama, Japan
•
A. M. Mandalari (UC3M), M. Bagnulo (UC3M), A. Lutu (Simula research). "TCP
Fast Open: initial measurements". To be submitted to PAM'16.
Anna Maria Mandalari, ESR 13 UC3M
Training & Inspiration
•
•
•
•
•
•
•
•
•
•
UC3M Master in Telematic Engineering
Statistics courses at UC3M
TPC Member at SWANSITY Workshop (IEEE SECON)
Seminar at UC3M “CDN, TLS, Tor, DNS Sec, Internet at the speed of
light, BGP” by Bruce Maggs
Presentation at Imdea Networks, July 2015, Madrid (Spain).
Presentation at SIGCOMM C2B(I)D Workshop, Aug 2015, London.
Mentoring session at SIGCOMM with Dave Clark (MIT) and Fahad
Dogar (Tufts University), Aug 2015, London.
Presentation at JNIC2015, July 2015, León (Spain).
Travel Grant to partecipate at IMC2015
Bachelor’s tesis Director, José Fernando Calcerrada (UC3M student)
Anna Maria Mandalari, ESR 13 UC3M
QoE: Case Studies
•
The case of the pervasive encryption
•
TCP Fast Open
•
HTTP/2
Anna Maria Mandalari, ESR 13 UC3M
Case study: TCP Fast Open
Send data directly in the SYN packet
J.Chu,A.Jain,Y.Cheng,andS.Radhakrishnan,“Tcpfastopen,”2014.
Anna Maria Mandalari, ESR 13 UC3M
Case study: TCP Fast Open
Middleboxes compatibility
Potential issues:
Major criticisme: Middleboxes might filter traffic that does not conform to expected
behaviors.
•
Middleboxes might not understand the TFO option and drop the packet;
•
Middleboxes might also drop packet with data in the initial SYN;
•
Carrier Grade NAT (CGN)
M. Honda et al., “Is it still possible to extend TCP?” in ACM IMC, 2011.
Anna Maria Mandalari, ESR 13 UC3M
QoE: Case Studies
•
The case of the pervasive encryption
•
TCP Fast Open
•
HTTP/2
Anna Maria Mandalari, ESR 13 UC3M
Case study: HTTP/2
From SPDY to HTTP/2
Anna Maria Mandalari, ESR 13 UC3M
Case study: HTTP/2. Middleboxes
compatibility
Does HTTP/2 require encryption?
Some implementations have stated that they will only support
HTTP/2 when it is used over an encrypted connection
(Google, Twitter…), and currently no browser supports HTTP/2
unencrypted!
Again: Middleboxes might filter traffic that does not conform to expected behaviors (HTTP proxies).
Anna Maria Mandalari, ESR 13 UC3M
Future career
Academia or Industry?
Internship in Telefonica Research
Anna Maria Mandalari, ESR 13 UC3M
STUDENT PRESENTATIONS
• Xiao Chen, ESR 11 UCL
• Supervisor: Prof. Marco Canini, Prof. Olivier
Bonaventure
Xiao Chen, ESR 11 UCL
My brief intro
• My name is Xiao Chen
• PhD student @ Université catholique de Louvain (UCL)
• Master degree obtained from European Master in
Distributed Computing (EMDC): IST, Portugal & KTH,
Sweden (07/2014)
• Internship (master thesis) @ Spotify
•
Efficient Distributed Pipelines for Anomaly Detection on Massive Production Logs
• Bachelor degree obtained from Dalian University of
Technology (China), Software Engineering
Xiao Chen, ESR 11 UCL
ESR11: Exposing measurement data to
the applications
• Université catholique de Louvain (UCL)
• Start from 04/11/2014
• Supervised by Prof. Marco Canini and Prof. Olivier
Bonaventure
• Work on WP3: APIs and Applications for Advanced
Network Monitoring
Xiao Chen, ESR 11 UCL
Main results obtained so far
• Contribute to D3.1: Requirements for API to expose
measurements;
• A scheduler system design for SDN control plane (in
progress);
Xiao Chen, ESR 11 UCL
Main results obtained so far
• Contribute to D3.1: Requirements for API to expose
measurements;
• Objective: Build a system to measure different (datacenter)
network’s performance.
• A scheduler system design for SDN control plane (in
progress);
Xiao Chen, ESR 11 UCL
Xiao Chen, ESR 11 UCL
Main results obtained so far
• Contribute to D3.1: Requirements for API to expose
measurements;
• A scheduler system design for SDN control plane (in
progress);
• Objective: To schedule tasks across SDN switches.
• What is the difference? relations?
Xiao Chen, ESR 11 UCL
Xiao Chen, ESR 11 UCL
Training:
• METRICS academic activities
• SDN summer school in Crete, Greece (July 2 - 3, 2015,
travel funded by METRICS)
• Present a poster @ Young Research Day
• “Passive Network Traffic Measurement” seminar
• FNRS networking group meeting
• Short-term lectures: Cloud Computing course, Distributed
Computing courses
• INGI LunchTime seminar, INL group meeting
• etc.
Xiao Chen, ESR 11 UCL
Hard technical points
• How to combine different sources of information to draw a
full picture of the network performance?
• How to monitor netwroks performance without heavily
effecting the original network?
• How to build a resilient, scalable distributed monitoring
framework?
• etc.
Xiao Chen, ESR 11 UCL
Possible solutions
• How to combine different sources of information to draw a
full picture of the network performance?
• Carefully design and implement the monitoring system by
selecting proper sources of information in different
context.
• How to monitor netwroks performance without heavily
effecting the original network?
• How to build a resilient, scalable distributed monitoring
framework?
• etc...
Xiao Chen, ESR 11 UCL
Possible solutions
• How to combine different sources of information to draw a
full picture of the network performance?
• How to monitor netwroks performance without heavily
effecting the original network?
• Passive monitoring methods can be used, but we still
need active monitoring methods.
• How to build a resilient, scalable distributed monitoring
framework?
• etc...
Xiao Chen, ESR 11 UCL
Possible solutions
• How to combine different sources of information to draw a
full picture of the network performance?
• How to monitor netwroks performance without heavily
effecting the original network?
• How to build a resilient, scalable distributed monitoring
framework?
• Apply fault tolerence mechanism, parallelization
techniques
• etc...
Xiao Chen, ESR 11 UCL
What’s next?
•
•
•
•
Refine and implement the proposed API;
Develop the system framework;
Investigate novel mechanisms;
Desirable improvements in the network applications:
• Achieving higher resource utilizations and more predictable
performance;
• Optimizing reaction times in response to failures;
• Being able to automatically troubleshoot performance issues of
distributed applications.
Xiao Chen, ESR 11 UCL
About my future career
• I would like to go to industry after my PhD;
• Target at a research department in a big company;
• Or join a cutting-edge tech start-up (maybe in China?) …
Xiao Chen, ESR 11 UCL
STUDENT PRESENTATIONS
• Viet Hoang TRAN, ESR 10 UCL
• Supervisor: Prof. Olivier Bonaventure,
Prof. Ramin Sadre
Viet Hoang TRAN, ESR 10 UCL
A brief intro about me
• I am Viet Hoang TRAN
• Hometown: Hanoi, Vietnam
• Education
I received B.E and Msc. in Computer
Networks
at Hanoi University of Science and
Technology
• Internship at NII, Japan (winter 2013)
Main topic is Network resilience, specifically
about Link Aggregation and Linux bonding
Viet Hoang TRAN, ESR 10 UCL
ESR 10: Understanding and modeling Multipath
TCP
Université catholique de Louvain (UCL)
Supervisor: Prof. Olivier Bonaventure,
Prof. Ramin Sadre.
Start: August 2014
source: www.etudiantsliberaux.be/temp/?p=1609
Viet Hoang TRAN, ESR 10 UCL
Viet Hoang TRAN, ESR 10 UCL
Multipath TCP in a nutshell
A significant TCP extension to aggregate multiple TCP
connections possibly over multiple interfaces, for:
• Goodput aggregation
• Handover
• Latency reduction
3G/4
G
Viet Hoang TRAN, ESR 10 UCL
Multipath TCP components
Path manager: establishes different paths between
interfaces of two hosts
Packet scheduler: selects which subflow that a packet
need to be sent through
Congestion control:
- As efficient as TCP’s congestion control
- But should be fair and friendly to regular TCP.
Viet Hoang TRAN, ESR 10 UCL
Approach
Modeling
Gather understanding
Passive
Measurements
Active
Measurements
Simulation/
Emulation
Validating
Model
Characterizing
MPTCP traffic
Predicting MPTCP
performance
Modeling
Tuning MPTCP
Viet Hoang TRAN, ESR 10 UCL
Progress
Measuring
• Collecting real MPTCP traffic at
multipath-tcp.org server from clients
around the world.
• about 134 GBytes/week
– ~ 20% of them are MPTCP
– ~ 6000 MPTCP connections
• Applied analyses on the dataset
Viet Hoang TRAN, ESR 10 UCL
Some Results
How current MPTCP creates subflows
How current MPTCP manages subflows
Performance overhead
Viet Hoang TRAN, ESR 10 UCL
Number of subflows per connection
Viet Hoang TRAN, ESR 10 UCL
How fast MPTCP establishes second
subflow
~ 90% of connections
create second subflow in
first second
~50% of connections
create second subflow in
100 ms
Viet Hoang TRAN, ESR 10 UCL
Round Trip Time difference between
subflows
MPTCP needs to select which subflow that a packet need
to be sent through.
Controlled by Scheduler, the default one works as follows:
1.among the subflows that have room inside their
congestion window
2.select the subflow with the lowest RTT
Viet Hoang TRAN, ESR 10 UCL
RTT difference can be large
Viet Hoang TRAN, ESR 10 UCL
RTT difference can be large
more than 1
second!
Viet Hoang TRAN, ESR 10 UCL
Data-resending overhead
2 issues:
TCP Retransmission:
resend of data over the
same subflow
MPTCP Reinjection: resend
of data over two or more
subflows
Viet Hoang TRAN, ESR 10 UCL
Data-resending overhead
2 issues:
TCP Retransmission:
resend of data over the
same subflow
MPTCP Reinjection: resend
of data over two or more
subflows
Viet Hoang TRAN, ESR 10 UCL
Progress - Workshop & Conferences
Security Summer school
July 13-17, 2015
April 21-24, 2015
Aug. 25 - Sept. 5, 2014
Bootcamp
Heidelberg
Industrial workshop
Feb. 5 - 6, 2015
TMA
AIMS, Gent
Ph.D workshop
June 22 - 25, 2015
Viet Hoang TRAN, ESR 10 UCL
Estonia
Progress - Publications &
Conferences
TMA workshop
AIMS Ph.D. workshop
• A first look at real Multipath TCP
traffic.
• Measuring and Modeling
• Hesmans, B., Tran Viet Hoang,
Sadre, R., and Bonaventure, O.
• Viet-Hoang Tran, Ramin Sadre,
• In Traffic Monitoring and
Analysis, vol. 9053 of LNCS.
• Acceptance rate: 9/24
• June 22 - 25, 2015
• Acceptance rate: 16/54 (29.6%)
• University of Ghent, Belgium
Multipath TCP.
and Olivier Bonaventure.
• April 21-24, 2015
• UPC, Barcelona, Spain
Viet Hoang TRAN, ESR 10 UCL
Technical Challenges
measurement
modeling
• Complexity of MPTCP itself
• Scalability
E.g.: Apple deployed MPTCP on
iOS (hundred millions of
clients!) but the dataset is not
public.
• Limited knowledge of remote
host: e.g.: MPTCP
implementations and
configurations
it has many components: path
manager, scheduler, congestion
control
• How MPTCP interacts with
other network factors?
– new middleboxes,
– topologies,
– delay difference between
paths
Viet Hoang TRAN, ESR 10 UCL
Next steps
Extend the measurements
larger dataset (5 months)
more extensive analyses
more dataset from Korea telecom
Develop models for MPTCP,
build simulations with ns3-dce
use real Linux MPTCP network stack and applications
the performance results not affected by hardware resource (happened with Mininet)
Viet Hoang TRAN, ESR 10 UCL
Modeling
• Developed simulations using ns3-dce to support
modeling
• Current focus: bufferbloat issue which is common today
• Experiment results: which is pretty inline with realistic
observation
Viet Hoang TRAN, ESR 10 UCL
Personal future career
I would like to continue academic and research career:
• Probably come back to HUST, Vietnam
• Co-operate with companies in networking domains
Viet Hoang TRAN, ESR 10 UCL