Advances in Internet data flow transport ISAE-SUPAERO Macquarie - - PowerPoint PPT Presentation

Advances in Internet data flow transport

ISAE-SUPAERO Macquarie University, the 22nd November 2018

ISAE-SUPAERO 2018 1 / 1

Toulouse complex for space and aeronautic Toulouse, south west of France

◮ 1 000 000 inhabitants and we love rugby union! ISAE-SUPAERO 2018 2 / 1

Toulouse complex for space and aeronautic Toulouse, south west of France

◮ 1 000 000 inhabitants and we love rugby union!

ISAE-SUPAERO

◮ top one school in aeronautical engineering under the ministry of

defense umbrella (application year for X students)

◮ higly selective access and cost free ◮ international masters (2 years after bachelor) 10/15KE per year ◮ Europe’s largest windtunnel for academic aeroacoustic research ◮ my team research deals with drone technology, satellite and deep space

communications, avionic systems, embedded systems, maths, ...

ISAE-SUPAERO 2018 2 / 1

Toulouse complex for space and aeronautic European Aeronautics and Space Pole

◮ Aeronautics : AIRBUS, ATR, SAFRAN, THALES, Liebherr Aerospace,

Rockwell Collins, Lat´ eco` ere, Air France

⋆ Manpower : 53 000 - including 5,000 researchers ⋆ Leading European centre for civil aviation industry ISAE-SUPAERO 2018 3 / 1

Toulouse complex for space and aeronautic European Aeronautics and Space Pole

◮ Aeronautics : AIRBUS, ATR, SAFRAN, THALES, Liebherr Aerospace,

Rockwell Collins, Lat´ eco` ere, Air France

⋆ Manpower : 53 000 - including 5,000 researchers ⋆ Leading European centre for civil aviation industry ◮ Space : THALES ALENIA SPACE, CNES, AIRBUS DEFENSE &

SPACE

⋆ Manpower : 12 000 (25% of the European manpower) ⋆ Leading European centre for satellites industry and earth observation ISAE-SUPAERO 2018 3 / 1

Toulouse complex for space and aeronautic European Aeronautics and Space Pole

◮ Aeronautics : AIRBUS, ATR, SAFRAN, THALES, Liebherr Aerospace,

Rockwell Collins, Lat´ eco` ere, Air France

⋆ Manpower : 53 000 - including 5,000 researchers ⋆ Leading European centre for civil aviation industry ◮ Space : THALES ALENIA SPACE, CNES, AIRBUS DEFENSE &

SPACE

⋆ Manpower : 12 000 (25% of the European manpower) ⋆ Leading European centre for satellites industry and earth observation

New Aerospace Valley international cluster B612 (aeronautical technologies and embedded systems) opened in april 2018

◮ 500 startups and middle sized companies hosted ISAE-SUPAERO 2018 3 / 1

Introduction The problem that takes most of my research time is data transfer

ISAE-SUPAERO 2018 4 / 1

Introduction The problem that takes most of my research time is data transfer Most of the time I need to provide transfer services that fit application needs in terms of reliability and delivery ratio in the most efficient and secure way

ISAE-SUPAERO 2018 4 / 1

Introduction The problem that takes most of my research time is data transfer Most of the time I need to provide transfer services that fit application needs in terms of reliability and delivery ratio in the most efficient and secure way

ISAE-SUPAERO 2018 4 / 1

Introduction The problem that takes most of my research time is data transfer Most of the time I need to provide transfer services that fit application needs in terms of reliability and delivery ratio in the most efficient and secure way

ISAE-SUPAERO 2018 4 / 1

Introduction The problem that takes most of my research time is data transfer Most of the time I need to provide transfer services that fit application needs in terms of reliability and delivery ratio in the most efficient and secure way

ISAE-SUPAERO 2018 4 / 1

Introduction The problem that takes most of my research time is data transfer Most of the time I need to provide transfer services that fit application needs in terms of reliability and delivery ratio in the most efficient and secure way

ISAE-SUPAERO 2018 4 / 1

Introduction The problem that takes most of my research time is data transfer Most of the time I need to provide transfer services that fit application needs in terms of reliability and delivery ratio in the most efficient and secure way

ISAE-SUPAERO 2018 4 / 1

Introduction The problem that takes most of my research time is data transfer Most of the time I need to provide transfer services that fit application needs in terms of reliability and delivery ratio in the most efficient and secure way

⇒ i.e. without deeply impacting current data traffic

ISAE-SUPAERO 2018 4 / 1

Introduction The problem that takes most of my research time is data transfer Most of the time I need to provide transfer services that fit application needs in terms of reliability and delivery ratio in the most efficient and secure way

⇒ i.e. without deeply impacting current data traffic

Basically my job is to fight against latency

ISAE-SUPAERO 2018 4 / 1

Introduction I can act on :

ISAE-SUPAERO 2018 5 / 1

Introduction I can act on :

◮ End-hosts, by proposing communication protocols or schemes to be

used conjointly within a communication protocol

⋆ At set of rules to control the sending of data and their reception ⋆ Mechanisms to recover missing data ISAE-SUPAERO 2018 5 / 1

Introduction I can act on :

◮ End-hosts, by proposing communication protocols or schemes to be

used conjointly within a communication protocol

⋆ At set of rules to control the sending of data and their reception ⋆ Mechanisms to recover missing data ◮ At the edge (mostly middleboxes, satellite gateway) or inside the core

network (if one day I work for a network vendor), by proposing Active Queue Management (AQM) or scheduling schemes

⋆ A set of rules to control admission, queuing, traffic policy, ... ⋆ Objective is to speedup traffic ISAE-SUPAERO 2018 5 / 1

Introduction I can act on :

◮ End-hosts, by proposing communication protocols or schemes to be

used conjointly within a communication protocol

⋆ At set of rules to control the sending of data and their reception ⋆ Mechanisms to recover missing data ◮ At the edge (mostly middleboxes, satellite gateway) or inside the core

network (if one day I work for a network vendor), by proposing Active Queue Management (AQM) or scheduling schemes

⋆ A set of rules to control admission, queuing, traffic policy, ... ⋆ Objective is to speedup traffic ◮ It is also possible to design mechanisms that act conjointly or in a

collaborative manner at the end-hosts and inside the network

ISAE-SUPAERO 2018 5 / 1

Where does it take place ? congestion control rate control admission control error control

ISAE-SUPAERO 2018 6 / 1

Where does it take place ? congestion control rate control admission control error control router/gateway scheduling/AQM/tunneling medium access control congestion control admission control error control

ISAE-SUPAERO 2018 6 / 1

Why doing research in this area Internet constantly evolves

ISAE-SUPAERO 2018 7 / 1

Why doing research in this area Internet constantly evolves blablabla... No way ! Are you kidding ?

ISAE-SUPAERO 2018 7 / 1

Why doing research in this area Internet constantly evolves blablabla... No way ! Are you kidding ? The truth is

ISAE-SUPAERO 2018 7 / 1

Why doing research in this area Internet constantly evolves blablabla... No way ! Are you kidding ? The truth is

◮ I love that, it’s fun ! ISAE-SUPAERO 2018 7 / 1

Why doing research in this area Internet constantly evolves blablabla... No way ! Are you kidding ? The truth is

◮ I love that, it’s fun ! ◮ I mostly get money from the satellite partners, aeonautical partners (@

is on board, embedded cameras too), broadcasting companies, video provider, ISP, ...

ISAE-SUPAERO 2018 7 / 1

Why doing research in this area Internet constantly evolves blablabla... No way ! Are you kidding ? The truth is

◮ I love that, it’s fun ! ◮ I mostly get money from the satellite partners, aeonautical partners (@

is on board, embedded cameras too), broadcasting companies, video provider, ISP, ...

ISAE-SUPAERO 2018 7 / 1

Why doing research in this area Internet constantly evolves blablabla... No way ! Are you kidding ? The truth is

◮ I love that, it’s fun ! ◮ I mostly get money from the satellite partners, aeonautical partners (@

is on board, embedded cameras too), broadcasting companies, video provider, ISP, ...

ISAE-SUPAERO 2018 7 / 1

Why doing research in this area Internet constantly evolves blablabla... No way ! Are you kidding ? The truth is

◮ I love that, it’s fun ! ◮ I mostly get money from the satellite partners, aeonautical partners (@

is on board, embedded cameras too), broadcasting companies, video provider, ISP, ...

◮ First question from them : ”How can I speed up my video flow/traffic

• r increase my QoE with your stuff ?”

ISAE-SUPAERO 2018 7 / 1

Why doing research in this area Internet constantly evolves blablabla... No way ! Are you kidding ? The truth is

◮ I love that, it’s fun ! ◮ I mostly get money from the satellite partners, aeonautical partners (@

is on board, embedded cameras too), broadcasting companies, video provider, ISP, ...

◮ First question from them : ”How can I speed up my video flow/traffic

• r increase my QoE with your stuff ?”

Trends are clear

◮ Huge video streaming traffic ⋆ France, corresponds to 90% of the Internet traffic since 2012 (IDATE) ⋆ By 2021, 80% of the world’s Internet traffic will we video (CISCO) ISAE-SUPAERO 2018 7 / 1

Why doing research in this area Internet constantly evolves blablabla... No way ! Are you kidding ? The truth is

◮ I love that, it’s fun ! ◮ I mostly get money from the satellite partners, aeonautical partners (@

is on board, embedded cameras too), broadcasting companies, video provider, ISP, ...

◮ First question from them : ”How can I speed up my video flow/traffic

• r increase my QoE with your stuff ?”

Trends are clear

◮ Huge video streaming traffic ⋆ France, corresponds to 90% of the Internet traffic since 2012 (IDATE) ⋆ By 2021, 80% of the world’s Internet traffic will we video (CISCO) ◮ Mobility : your mobile is your computer ⋆ Mobile internet traffic as percentage of total web traffic : 20% (2012),

40% (2015), 52% in January 2018

⋆ Huge for some countries : India 79%, Singapore 78% 1

• 1. https://www.statista.com/statistics/430830/share-of-mobile-internet-traffic-countries/

ISAE-SUPAERO 2018 7 / 1

So OK, Internet contantly evolves Problem : past hypothesis used to build communication protocols do not hold today (the truth since 1960 !) How to tackle these evolutions ? Common method is to propose Yet Another Communication Protocol able to handle ⇒ application needs ⇒ new media and communication paradigms

ISAE-SUPAERO 2018 8 / 1

So OK, Internet contantly evolves Problem : past hypothesis used to build communication protocols do not hold today (the truth since 1960 !) How to tackle these evolutions ? Common method is to propose Yet Another Communication Protocol able to handle ⇒ application needs ⇒ new media and communication paradigms In brief nothing new and this is what I do But this approach still holds ?

ISAE-SUPAERO 2018 8 / 1

Non-exhaustive congestion control mechanisms timeline End-to-end protocols Core network sched/AQM Tahoe GPS LAS WFQ 1 9 9

ISAE-SUPAERO 2018 9 / 1

Non-exhaustive congestion control mechanisms timeline End-to-end protocols Core network sched/AQM Tahoe GPS LAS WFQ 1 9 9 Reno Newreno Vegas SACK RED BLUE ECN CHOKE WF2Q+ 2

ISAE-SUPAERO 2018 9 / 1

Non-exhaustive congestion control mechanisms timeline End-to-end protocols Core network sched/AQM Tahoe GPS LAS WFQ 1 9 9 Reno Newreno Vegas SACK RED BLUE ECN CHOKE WF2Q+ 2 Westwood DCCP Skype Hybla SCTP Westwood+ HTCP FAST Compound VENO BIC CUBIC YELLOW XCP A-RED RIO A-RIO RUN2C DCTCP 2 1

ISAE-SUPAERO 2018 9 / 1

Non-exhaustive congestion control mechanisms timeline End-to-end protocols Core network sched/AQM Tahoe GPS LAS WFQ 1 9 9 Reno Newreno Vegas SACK RED BLUE ECN CHOKE WF2Q+ 2 Westwood DCCP Skype Hybla SCTP Westwood+ HTCP FAST Compound VENO BIC CUBIC YELLOW XCP A-RED RIO A-RIO RUN2C DCTCP 2 1 QUIC LEDBAT MPTCP BBR SPROUT REMY WebRTC CODEL PIE pFABRIC CBACC

ISAE-SUPAERO 2018 9 / 1

Where these solutions take place ? Applications user-level kernel-level Operating System hardware Applications

ISAE-SUPAERO 2018 10 / 1

Where these solutions take place ? Applications user-level kernel-level Operating System hardware Applications Transport

ISAE-SUPAERO 2018 10 / 1

Where these solutions take place ? Applications user-level kernel-level Operating System hardware Applications Transport Network

ISAE-SUPAERO 2018 10 / 1

Where these solutions take place ? Applications user-level kernel-level Operating System hardware Applications Transport Network Link 011010110100101

ISAE-SUPAERO 2018 10 / 1

Where these solutions take place ? Applications user-level kernel-level Operating System hardware Applications TCP UDP ... Network Link 011010110100101

ISAE-SUPAERO 2018 10 / 1

Where these solutions take place ? Applications user-level kernel-level Operating System hardware Applications TCP UDP ... You are here ! Network Link 011010110100101

ISAE-SUPAERO 2018 10 / 1

Where these solutions take place ? Applications user-level kernel-level Operating System hardware Applications TCP UDP ... AQM SCHED ... Link 011010110100101

ISAE-SUPAERO 2018 10 / 1

Where these solutions take place ? Applications user-level kernel-level Operating System hardware Applications TCP UDP ... AQM SCHED ... You are here ! Link 011010110100101

ISAE-SUPAERO 2018 10 / 1

End-to-end protocols deployment Main problem : you must convince OS vendors, GNU/Linux consortium, same problem with layer 3 mechanisms but a bit more accessible

ISAE-SUPAERO 2018 11 / 1

End-to-end protocols deployment Main problem : you must convince OS vendors, GNU/Linux consortium, same problem with layer 3 mechanisms but a bit more accessible Hypothesis : your idea is a good/great idea

ISAE-SUPAERO 2018 11 / 1

End-to-end protocols deployment Main problem : you must convince OS vendors, GNU/Linux consortium, same problem with layer 3 mechanisms but a bit more accessible Hypothesis : your idea is a good/great idea Step #1 enthusiastic - you publish, communicate, run over standardisation bodies (IEEE, ITU-T, IETF, ...)

ISAE-SUPAERO 2018 11 / 1

End-to-end protocols deployment Main problem : you must convince OS vendors, GNU/Linux consortium, same problem with layer 3 mechanisms but a bit more accessible Hypothesis : your idea is a good/great idea Step #1 enthusiastic - you publish, communicate, run over standardisation bodies (IEEE, ITU-T, IETF, ...)

”I will be the next Van Jacobson of the transport layer !”

ISAE-SUPAERO 2018 11 / 1

End-to-end protocols deployment Main problem : you must convince OS vendors, GNU/Linux consortium, same problem with layer 3 mechanisms but a bit more accessible Hypothesis : your idea is a good/great idea Step #1 enthusiastic - you publish, communicate, run over standardisation bodies (IEEE, ITU-T, IETF, ...)

”I will be the next Van Jacobson of the transport layer !”

Step #2 frustration - ”Why my fantastic protocol is not accepted !”

ISAE-SUPAERO 2018 11 / 1

End-to-end protocols deployment Main problem : you must convince OS vendors, GNU/Linux consortium, same problem with layer 3 mechanisms but a bit more accessible Hypothesis : your idea is a good/great idea Step #1 enthusiastic - you publish, communicate, run over standardisation bodies (IEEE, ITU-T, IETF, ...)

”I will be the next Van Jacobson of the transport layer !”

Step #2 frustration - ”Why my fantastic protocol is not accepted !” Step #3 resignation - ”They don’t know what they are lossing. I give up I will limit myself to publish papers.”

ISAE-SUPAERO 2018 11 / 1

End-to-end protocols deployment Main problem : you must convince OS vendors, GNU/Linux consortium, same problem with layer 3 mechanisms but a bit more accessible Hypothesis : your idea is a good/great idea Step #1 enthusiastic - you publish, communicate, run over standardisation bodies (IEEE, ITU-T, IETF, ...)

”I will be the next Van Jacobson of the transport layer !”

Step #2 frustration - ”Why my fantastic protocol is not accepted !” Step #3 resignation - ”They don’t know what they are lossing. I give up I will limit myself to publish papers.” Step #4 doubt (moreover may be after paper rejection) - ”Am I a bad researcher ? Did I miss something ?”

ISAE-SUPAERO 2018 11 / 1

End-to-end protocols deployment Main problem : you must convince OS vendors, GNU/Linux consortium, same problem with layer 3 mechanisms but a bit more accessible Hypothesis : your idea is a good/great idea Step #1 enthusiastic - you publish, communicate, run over standardisation bodies (IEEE, ITU-T, IETF, ...)

”I will be the next Van Jacobson of the transport layer !”

Step #2 frustration - ”Why my fantastic protocol is not accepted !” Step #3 resignation - ”They don’t know what they are lossing. I give up I will limit myself to publish papers.” Step #4 doubt (moreover may be after paper rejection) - ”Am I a bad researcher ? Did I miss something ?” Step #5 therapy - ”You are not alone...

ISAE-SUPAERO 2018 11 / 1

Bad timing theory David Clark (MIT) ”apocalypse of the two elephants” theory on the OSI model failure against DoD one (i.e. TCP/IP model)

ISAE-SUPAERO 2018 12 / 1

Bad timing theory David Clark (MIT) ”apocalypse of the two elephants” theory on the OSI model failure against DoD one (i.e. TCP/IP model) Activity Time Research Standards

ISAE-SUPAERO 2018 12 / 1

Bad timing theory David Clark (MIT) ”apocalypse of the two elephants” theory on the OSI model failure against DoD one (i.e. TCP/IP model) Activity Time Research Standards

The challenge is to come up with standards between both ”elephants”

ISAE-SUPAERO 2018 12 / 1

Bad timing theory David Clark (MIT) ”apocalypse of the two elephants” theory on the OSI model failure against DoD one (i.e. TCP/IP model) Activity Time Research Standards

The challenge is to come up with standards between both ”elephants” Before ⇒ not well understood, there is on-going research

ISAE-SUPAERO 2018 12 / 1

Bad timing theory David Clark (MIT) ”apocalypse of the two elephants” theory on the OSI model failure against DoD one (i.e. TCP/IP model) Activity Time Research Standards

The challenge is to come up with standards between both ”elephants” Before ⇒ not well understood, there is on-going research After ⇒ investment done, too late mate !

ISAE-SUPAERO 2018 12 / 1

How to thwart slow new proposals adoption Applications user-level kernel-level Operating System hardware

ISAE-SUPAERO 2018 13 / 1

How to thwart slow new proposals adoption Applications user-level kernel-level Operating System hardware Applications

ISAE-SUPAERO 2018 13 / 1

How to thwart slow new proposals adoption Applications user-level kernel-level Operating System hardware Applications UL Transport

ISAE-SUPAERO 2018 13 / 1

How to thwart slow new proposals adoption Applications user-level kernel-level Operating System hardware Applications UL Transport Transport Network Link 011010110100101

ISAE-SUPAERO 2018 13 / 1

How to thwart slow new proposals adoption Applications user-level kernel-level Operating System hardware Applications HTTP, DASH, ... TCP Network Link 011010110100101

ISAE-SUPAERO 2018 13 / 1

How to thwart slow new proposals adoption Applications user-level kernel-level Operating System hardware Applications RTP UDP Network Link 011010110100101

ISAE-SUPAERO 2018 13 / 1

How to thwart slow new proposals adoption Applications user-level kernel-level Operating System hardware Applications QUIC UDP Network Link 011010110100101

ISAE-SUPAERO 2018 13 / 1

How to thwart slow new proposals adoption Applications user-level kernel-level Operating System hardware Applications QUIC The messie protocol ? UDP Network Link 011010110100101

ISAE-SUPAERO 2018 13 / 1

Proposing a new protocol You MUST prevent congestion (for the network)

◮ Does not mean you are mandatory TCP-friendly see Flow Rate

Fairness : Dismantling a Religion by Bob Briscoe

ISAE-SUPAERO 2018 14 / 1

Proposing a new protocol You MUST prevent congestion (for the network)

◮ Does not mean you are mandatory TCP-friendly see Flow Rate

Fairness : Dismantling a Religion by Bob Briscoe

◮ So congestion occurs when the traffic increases ISAE-SUPAERO 2018 14 / 1

Proposing a new protocol You MUST prevent congestion (for the network)

◮ Does not mean you are mandatory TCP-friendly see Flow Rate

Fairness : Dismantling a Religion by Bob Briscoe

◮ So congestion occurs when the traffic increases resulting in a slowing

down of the QoE

ISAE-SUPAERO 2018 14 / 1

Proposing a new protocol You MUST prevent congestion (for the network)

◮ Does not mean you are mandatory TCP-friendly see Flow Rate

Fairness : Dismantling a Religion by Bob Briscoe

◮ So congestion occurs when the traffic increases resulting in a slowing

down of the QoE

You SHOULD care about privacy (for the user)

ISAE-SUPAERO 2018 14 / 1

Proposing a new protocol You MUST prevent congestion (for the network)

◮ Does not mean you are mandatory TCP-friendly see Flow Rate

Fairness : Dismantling a Religion by Bob Briscoe

◮ So congestion occurs when the traffic increases resulting in a slowing

down of the QoE

You SHOULD care about privacy (for the user) To do so, we use collaborating congestion control algorithms that can be placed anywhere in the network

ISAE-SUPAERO 2018 14 / 1

Proposing a new protocol You MUST prevent congestion (for the network)

◮ Does not mean you are mandatory TCP-friendly see Flow Rate

Fairness : Dismantling a Religion by Bob Briscoe

◮ So congestion occurs when the traffic increases resulting in a slowing

down of the QoE

You SHOULD care about privacy (for the user) To do so, we use collaborating congestion control algorithms that can be placed anywhere in the network That’s enough ? NO ⇒ your protocol must be compliant with network operator policies

ISAE-SUPAERO 2018 14 / 1

The QUIC protocol QUIC (Quick UDP Internet Connections) is a new encrypted-by-default Internet transport protocol proposed by Google with the following characteristics :

◮ Version negotiation ◮ Low-latency connection establishment ◮ Authenticated and encrypted header and payload ◮ Stream multiplexing ◮ Stream and connection-level flow control ◮ Connection migration and resilience to NAT rebinding

Basically objectives are :

◮ to accelerate HTTP traffic and make it more secure ◮ ultimate goal is to replace TCP and TLS

On-going IETF standardisation Numerous implementations : Google-QUIC, PicoQuic (C. Huitema), IETF-QUIC, ...

ISAE-SUPAERO 2018 15 / 1

The QUIC case over SATCOM

• Need to protect from losses

High Delay x BW product Losses might occur Need to reach optimal capacity PEP sender receiver

SATCOM systems splits end-to-end connection

ISAE-SUPAERO 2018 16 / 1

The QUIC case over SATCOM

• Need to protect from losses

High Delay x BW product Losses might occur Need to reach optimal capacity PEP sender receiver

SATCOM systems splits end-to-end connection and they are not the only one

ISAE-SUPAERO 2018 16 / 1

QUIC over SATCOM vs 4G Satcom

ISAE-SUPAERO 2018 17 / 1

QUIC over SATCOM vs 4G Satcom

ISAE-SUPAERO 2018 17 / 1

QUIC over SATCOM vs 4G Satcom 4G

ISAE-SUPAERO 2018 17 / 1

QUIC over SATCOM vs 4G Satcom 4G

ISAE-SUPAERO 2018 17 / 1

To conclude with QUIC experiments QUIC is not better than splitted-TCP for a SATCOM public access

◮ At least for large files, splitted-TCP wins ◮ Further details : https://arxiv.org/abs/1810.04970

We believe satellite is not the only link impacted

◮ Want to see if you are concerned by this issue ? ◮ Download our code :

https://forge.net4sat.org/cnes/quxa-public

ISAE-SUPAERO 2018 18 / 1

Conclusion on this part TCP and UDP are enough at transport layer I believe the right approach is UL transport for QoE There is no one-fit-all solution

ISAE-SUPAERO 2018 19 / 1

Contributions

ISAE-SUPAERO 2018 20 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS DiffServ AQM for AF class

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED AQM based on Kohonen NN

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED 2011 ECN* Statistical method to estimate congestion

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED 2011 ECN* 2014 FaQ Speeding up TCP traffic

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED 2011 ECN* 2014 FaQ 2018 PSS Scheduling schemes, ongoing IETF work

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED 2011 ECN* 2014 FaQ 2018 PSS 2008 CL-WLAN Cross-layer scheme for WLAN

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED 2011 ECN* 2014 FaQ 2018 PSS 2008 CL-WLAN 2014 CLIFT Cross-layer scheme for satellite link

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED 2011 ECN* 2014 FaQ 2018 PSS 2008 CL-WLAN 2014 CLIFT 2012 DQN Routing protocol for quasi- deterministic LEO constellation with non-persistant paths

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED 2011 ECN* 2014 FaQ 2018 PSS 2008 CL-WLAN 2014 CLIFT 2012 DQN 2015 TRW Broadcast protocol based on Tempo- ral Random Walk for opportunistic networks

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED 2011 ECN* 2014 FaQ 2018 PSS 2008 CL-WLAN 2014 CLIFT 2012 DQN 2015 TRW 2006 gTFRC Transport protocols for QoS architec- ture

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED 2011 ECN* 2014 FaQ 2018 PSS 2008 CL-WLAN 2014 CLIFT 2012 DQN 2015 TRW 2006 gTFRC Transport protocols for QoS architec- ture 2008 Chameleon

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED 2011 ECN* 2014 FaQ 2018 PSS 2008 CL-WLAN 2014 CLIFT 2012 DQN 2015 TRW 2006 gTFRC 2008 Chameleon 2009 DCCP-sat Transport protocol for SAT- COM systems

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED 2011 ECN* 2014 FaQ 2018 PSS 2008 CL-WLAN 2014 CLIFT 2012 DQN 2015 TRW 2006 gTFRC 2008 Chameleon 2009 DCCP-sat Transport protocol for SAT- COM systems 2010 IP-ERN

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED 2011 ECN* 2014 FaQ 2018 PSS 2008 CL-WLAN 2014 CLIFT 2012 DQN 2015 TRW 2006 gTFRC 2008 Chameleon 2009 DCCP-sat Transport protocol for SAT- COM systems 2010 IP-ERN 2013 TCP-aero

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED 2011 ECN* 2014 FaQ 2018 PSS 2008 CL-WLAN 2014 CLIFT 2012 DQN 2015 TRW 2006 gTFRC 2008 Chameleon 2009 DCCP-sat 2010 IP-ERN 2013 TCP-aero 2013 eCMT-SCTP Transport protocol for multimedia, multipath, LBE, ...

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED 2011 ECN* 2014 FaQ 2018 PSS 2008 CL-WLAN 2014 CLIFT 2012 DQN 2015 TRW 2006 gTFRC 2008 Chameleon 2009 DCCP-sat 2010 IP-ERN 2013 TCP-aero 2013 eCMT-SCTP Transport protocol for multimedia, multipath, LBE, ... 2013 Q-AIMD

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED 2011 ECN* 2014 FaQ 2018 PSS 2008 CL-WLAN 2014 CLIFT 2012 DQN 2015 TRW 2006 gTFRC 2008 Chameleon 2009 DCCP-sat 2010 IP-ERN 2013 TCP-aero 2013 eCMT-SCTP Transport protocol for multimedia, multipath, LBE, ... 2013 Q-AIMD 2014 DAPS

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED 2011 ECN* 2014 FaQ 2018 PSS 2008 CL-WLAN 2014 CLIFT 2012 DQN 2015 TRW 2006 gTFRC 2008 Chameleon 2009 DCCP-sat 2010 IP-ERN 2013 TCP-aero 2013 eCMT-SCTP Transport protocol for multimedia, multipath, LBE, ... 2013 Q-AIMD 2014 DAPS 2015 FLOWER

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2004 APS 2008 K-RED 2011 ECN* 2014 FaQ 2018 PSS 2008 CL-WLAN 2014 CLIFT 2012 DQN 2015 TRW 2006 gTFRC 2008 Chameleon 2009 DCCP-sat 2010 IP-ERN 2013 TCP-aero 2013 eCMT-SCTP 2013 Q-AIMD 2014 DAPS 2015 FLOWER 2007-now Tetrys A-Tetrys, REST, TENTET,...

ISAE-SUPAERO 2018 21 / 1

Contributions Applications UL Transport Transport Network Uni/multicast DTN 2014 FaQ 2015 FLOWER 2007-now Tetrys A-Tetrys, REST, TENTET,...

ISAE-SUPAERO 2018 22 / 1

FavorQueue : a Parameterless Active Queue Management to Improve TCP Traffic Performance

ISAE-SUPAERO 2018 23 / 1

Latency matters i

0 1 2 3 4 5 6 0.001 0.01 0.1 1 Probability Density Function Flow duration [s] server client 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1 10 100 1000 Cumulative Distribution Function Flow length, L [pkt] server client

By courstesy of D. Ciullo, M. Mellia, and M. Meo. Two schemes to reduce latency in short lived TCP flows. Communications Letters, 13(10), October 2009.

ISAE-SUPAERO 2018 24 / 1

Latency matters i

0 1 2 3 4 5 6 0.001 0.01 0.1 1 Probability Density Function Flow duration [s] server client 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1 10 100 1000 Cumulative Distribution Function Flow length, L [pkt] server client

By courstesy of D. Ciullo, M. Mellia, and M. Meo. Two schemes to reduce latency in short lived TCP flows. Communications Letters, 13(10), October 2009.

ISAE-SUPAERO 2018 24 / 1

The facts More than 95% (70%) of the client (server) flows are shorter than 10 full-size segments As a result, their delay performance is mainly driven by :

◮ The end-to-end transfer delay ⋆ This delay can be reduced if the queueing delay of each router is low ◮ The potential losses at the beginning connection ⋆ The first packets lost at the beginning of a TCP connection (i.e. in the

slow-start phase) are mainly recovered by the RTO mechanism

⋆ At the beginning the initila value is(was) set to 1(3) second(s) ISAE-SUPAERO 2018 25 / 1

TCP loss detection scheme

ISAE-SUPAERO 2018 26 / 1

TCP loss detection scheme pkt#1

ISAE-SUPAERO 2018 26 / 1

TCP loss detection scheme pkt#1 ack#1

ISAE-SUPAERO 2018 26 / 1

TCP loss detection scheme pkt#1 ack#1 pkt#2 X

ISAE-SUPAERO 2018 26 / 1

TCP loss detection scheme pkt#1 ack#1 pkt#2 X pkt#3

ISAE-SUPAERO 2018 26 / 1

TCP loss detection scheme pkt#1 ack#1 pkt#2 X pkt#3 ack#1

ISAE-SUPAERO 2018 26 / 1

TCP loss detection scheme pkt#1 ack#1 pkt#2 X pkt#3 ack#1 pkt#4

ISAE-SUPAERO 2018 26 / 1

TCP loss detection scheme pkt#1 ack#1 pkt#2 X pkt#3 ack#1 pkt#4 ack#1

ISAE-SUPAERO 2018 26 / 1

TCP loss detection scheme pkt#1 ack#1 pkt#2 X pkt#3 ack#1 pkt#4 ack#1 pkt#5

ISAE-SUPAERO 2018 26 / 1

TCP loss detection scheme pkt#1 ack#1 pkt#2 X pkt#3 ack#1 pkt#4 ack#1 pkt#5 ack#1

ISAE-SUPAERO 2018 26 / 1

TCP loss detection scheme pkt#1 ack#1 pkt#2 X pkt#3 ack#1 pkt#4 ack#1 pkt#5 ack#1 pkt#6

ISAE-SUPAERO 2018 26 / 1

TCP loss detection scheme pkt#1 ack#1 pkt#2 X pkt#3 ack#1 pkt#4 ack#1 pkt#5 ack#1 pkt#6 pkt#7

ISAE-SUPAERO 2018 26 / 1

TCP loss detection scheme pkt#1 ack#1 pkt#2 X pkt#3 ack#1 pkt#4 ack#1 pkt#5 ack#1 pkt#6 pkt#7 pkt#8

ISAE-SUPAERO 2018 26 / 1

TCP loss detection scheme pkt#1 ack#1 pkt#2 X pkt#3 ack#1 pkt#4 ack#1 pkt#5 ack#1 pkt#6 pkt#7 pkt#8 pkt#2 3 DUP ACK

ISAE-SUPAERO 2018 26 / 1

TCP loss detection scheme pkt#1 ack#1 pkt#2 X pkt#3 ack#1 pkt#4 ack#1 pkt#5 ack#1 pkt#6 pkt#7 pkt#8 pkt#2 pkt#9

ISAE-SUPAERO 2018 26 / 1

Short flow loss detection problem

ISAE-SUPAERO 2018 27 / 1

Short flow loss detection problem pkt#1

ISAE-SUPAERO 2018 27 / 1

Short flow loss detection problem pkt#1 ack#1

ISAE-SUPAERO 2018 27 / 1

Short flow loss detection problem pkt#1 ack#1 pkt#2 X

ISAE-SUPAERO 2018 27 / 1

Short flow loss detection problem pkt#1 ack#1 pkt#2 X pkt#3

ISAE-SUPAERO 2018 27 / 1

Short flow loss detection problem pkt#1 ack#1 pkt#2 X pkt#3 ack#1

ISAE-SUPAERO 2018 27 / 1

Short flow loss detection problem pkt#1 ack#1 pkt#2 X pkt#3 ack#1 Timeout

ISAE-SUPAERO 2018 27 / 1

Short flow loss detection problem pkt#1 ack#1 pkt#2 X pkt#3 ack#1 Timeout pkt#2

ISAE-SUPAERO 2018 27 / 1

Short flow loss detection problem pkt#1 ack#1 pkt#2 X pkt#3 ack#1 Timeout pkt#2 BREAKING NEWS RACK has repla- ced all these detection schemes !

ISAE-SUPAERO 2018 27 / 1

Our proposal FavorQueue (FaQ) FavorQueue favors certain packets in order to speed up delay transfer by giving preferential access and to protect them from drop Follow a well-known queueing theory result : ”the overall mean latency is reduced when shortest job is served first” (L. Kleinrock Queuing Systems) Basic principle :

◮ When a packet is enqueued, a check is done on the whole queue to

seek another packet from the same flow

◮ If no other packet is found, it becomes a favored packet and is served

first

◮ Push-out when queue is full

FaQ stateless, no parameters to set

ISAE-SUPAERO 2018 28 / 1

Simulation experiment

0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1 10 100 1000 Cumulative Distribution Function Flow length (pkt) 1 2 3 4 5 6 10−6 10−5 10−4 10−3 10−2 10−1 Probability Density Function Latency (s) DropTail FavorQueue ISAE-SUPAERO 2018 29 / 1

Latency

0.00 0.50 1.00 1.50 2.00 2.50 0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.95 Latency (s) Load ρ DropTail Choke Run2C FavourQueue LAS

ISAE-SUPAERO 2018 30 / 1

Latency

0.1 1 10 100 1 10 100 1000 Latency (s) Flow length (pkt) DropTail FavourQueue

ISAE-SUPAERO 2018 31 / 1

Conclusion FaQ Presentation at the 87th IETF in Berlin (July 2013) In all experiments FaQ improves the latency up to 30% Stateless, no parameters Beneficial to all kind of TCP flows Code realeased within a TUN/TAP device

FavorQueue : A parameterless active queue management to improve TCP traffic performance, Pascal Anelli, R´ emi Diana, Emmanuel Lochin, (2014) Computer Networks, 60. 171-186. ISSN 1389-1286.

ISAE-SUPAERO 2018 32 / 1

FLOWER, a Fuzzy LOWER than Best-EffoRt Protocol

ISAE-SUPAERO 2018 33 / 1

Motivation Work with the french space agency (CNES) and Thales Alenia Space The initial question was : can we manage to fetch the unused satellite link capacity to :

◮ Optimize data link capacity over shared links ◮ Add a signaling or non-commercial traffic ◮ Provide Internet access for isolated area ◮ Provide a new service for new business

And all this without disturbing commercial traffic and in a E2E fashion ?

ISAE-SUPAERO 2018 34 / 1

Motivation Work with the french space agency (CNES) and Thales Alenia Space The initial question was : can we manage to fetch the unused satellite link capacity to :

◮ Optimize data link capacity over shared links ◮ Add a signaling or non-commercial traffic ◮ Provide Internet access for isolated area ◮ Provide a new service for new business

And all this without disturbing commercial traffic and in a E2E fashion ? Idea : let’s use an LBE protocol such as LEDBAT

ISAE-SUPAERO 2018 34 / 1

Motivation Work with the french space agency (CNES) and Thales Alenia Space The initial question was : can we manage to fetch the unused satellite link capacity to :

◮ Optimize data link capacity over shared links ◮ Add a signaling or non-commercial traffic ◮ Provide Internet access for isolated area ◮ Provide a new service for new business

And all this without disturbing commercial traffic and in a E2E fashion ? Idea : let’s use an LBE protocol such as LEDBAT and let’s do charity business

ISAE-SUPAERO 2018 34 / 1

LBE in a nutshell #pkts Time TCP

ISAE-SUPAERO 2018 35 / 1

LBE in a nutshell #pkts Time TCP LBE

ISAE-SUPAERO 2018 35 / 1

LBE in a nutshell #pkts Time TCP LBE TCP+LBE

ISAE-SUPAERO 2018 35 / 1

Problem LEDBAT does not have an optimal configuration and needs to be tuned as a function of the network state

◮ First attempt in a previous work where we found ”working regions” but

no generic/optimal configuration

LEDBAT is not adaptive and implements a P-type controller to control the sending rate as a function of the delay. Both parameters (target and gain) are fixed One possible solution would be to implement an adaptive gain inside this P-type controller but such adaptation would require a fine grained analytical model

◮ target remains set/chosen by the administrator and should not be

adaptive

ISAE-SUPAERO 2018 36 / 1

Bad LEDBAT behaviour

ISAE-SUPAERO 2018 37 / 1

Proposal The idea is to overtake this difficulty by using a fuzzy algorithm to perform such task Our proposal named FLOWER

◮ Aims at proposing a novel LBE transport protocol based on a fuzzy

logic algorithm

◮ Aims at reacting better than LEDBAT front to congestion event

Why Fuzzy Logic ?

◮ Allows to incorporate our heuristic knowledge about how to control

the system

◮ Low computational complexity ISAE-SUPAERO 2018 38 / 1

Correct behaviour

ISAE-SUPAERO 2018 39 / 1

Conclusion Presentation at the 93th IETF in Prague (July 2015) FLOWER : first solution that solves LEDBAT issues Implementation in Linux kernel available

FLOWER, an Innovative Fuzzy Lower-than-Best-Effort Transport Protocol, Si Quoc Viet Trang and Emmanuel Lochin, (2016) Computer Networks, 110. 18-30. ISSN 1389-1286.

ISAE-SUPAERO 2018 40 / 1

TETRYS and SPEERYT startup

http://www.speeryt.fr/

ISAE-SUPAERO 2018 41 / 1

At the beginning A novel reliable mechanism, based on an on-the-fly coding scheme Idea : combining both world of network coding and transport protocol In its most general form, full reliability can be achieved While being tolerant to loss of feedbacks Present good properties for realtime interactive and bidirectional apps (VoIP, videoconferencing) No self-clocked mechanism ⇒ Low data recovery delay

ISAE-SUPAERO 2018 42 / 1

There is more Real time transmissions Video-conferencing, drone communications Bulk Data Transfer Opportunistic Networking and Delay Tolerant Networking IP tunnelling Reliable multicast Satellite communications

ISAE-SUPAERO 2018 43 / 1

Basic principle

ISAE-SUPAERO 2018 44 / 1

Basic principle P1 P1

ISAE-SUPAERO 2018 44 / 1

Basic principle P1 P1 P2 X

ISAE-SUPAERO 2018 44 / 1

Basic principle P1 P1 P2 X R2

1

P2 R2

1 = α P1 ⊕ β P2

ISAE-SUPAERO 2018 44 / 1

Basic principle P1 P1 P2 X R2

1

P2

sack#2

X P3 X

ISAE-SUPAERO 2018 44 / 1

Basic principle P1 P1 P2 X R2

1

P2

sack#2

X P3 X P4 X

ISAE-SUPAERO 2018 44 / 1

Basic principle P1 P1 P2 X R2

1

P2

sack#2

X P3 X P4 X R4

1

X R4

1 = ⊕4 i=1αi Pi

ISAE-SUPAERO 2018 44 / 1

Basic principle P1 P1 P2 X R2

1

P2

sack#2

X P3 X P4 X R4

1

X P5 P5

ISAE-SUPAERO 2018 44 / 1

Basic principle P1 P1 P2 X R2

1

P2

sack#2

X P3 X P4 X R4

1

X P5 P5 P6 P6

ISAE-SUPAERO 2018 44 / 1

Basic principle P1 P1 P2 X R2

1

P2

sack#2

X P3 X P4 X R4

1

X P5 P5 P6 P6 R6

1

R6

1

ISAE-SUPAERO 2018 44 / 1

Basic principle P1 P1 P2 X R2

1

P2

sack#2

X P3 X P4 X R4

1

X P5 P5 P6 P6 R6

1

R6

1

P7 P7

ISAE-SUPAERO 2018 44 / 1

Basic principle P1 P1 P2 X R2

1

P2

sack#2

X P3 X P4 X R4

1

X P5 P5 P6 P6 R6

1

R6

1

P7 P7 P8 P8

ISAE-SUPAERO 2018 44 / 1

Basic principle P1 P1 P2 X R2

1

P2

sack#2

X P3 X P4 X R4

1

X P5 P5 P6 P6 R6

1

R6

1

P7 P7 P8 P8 R8

1

P3P4 R8

1

(R6

1, R8 1)T = G.(P3, P4)T

G =

• α(1..6)

3

α(1..6)

4

α(1..8)

3

α(1..8)

4

• sack#8

ISAE-SUPAERO 2018 44 / 1

Basic principle P1 P1 P2 X R2

1

P2

sack#2

X P3 X P4 X R4

1

X P5 P5 P6 P6 R6

1

R6

1

P7 P7 P8 P8 R8

1

P3P4 R8

1

sack#8

P9 P9

ISAE-SUPAERO 2018 44 / 1

Basic principle P1 P1 P2 X R2

1

P2

sack#2

X P3 X P4 X R4

1

X P5 P5 P6 P6 R6

1

R6

1

P7 P7 P8 P8 R8

1

P3P4 R8

1

sack#8

P9 P9 P10 P10

ISAE-SUPAERO 2018 44 / 1

Basic principle P1 P2 X R2

1

sack#2

X P3 X P4 X R4

1

X P5 P6 R6

1

P7 P8 R8

1

sack#8

P9 P9 P10 P10 R10

9

ISAE-SUPAERO 2018 44 / 1

Tetrys demo

https://www.youtube.com/watch?v=GxJzQKMJKZQ https://www.youtube.com/watch?v=wmPvEHfSUqQ

ISAE-SUPAERO 2018 45 / 1

Speeryt promotes the Tetrys library Powerful API using callbacks system Tetrys automatically ”sends” encoded/decoded packets to the encoder/decoder The source packets can have a variable size (with a maximum of 65536 bytes) Easy to integrate Tetrys inside existing code Easy to use for demo with commercial apps Can get some [en—de]coding statistics througth the callbacks by catching special events

On-the-fly erasure coding for real-time video applications, Pierre-Ugo Tournoux, Emmanuel Lochin, J´ erˆ

• me Lacan, Amine Bouabdallah, Vincent Roca,

(2011) IEEE Transactions on Multimedia, 13 (4). 797-812. ISSN 1520-9210

ISAE-SUPAERO 2018 46 / 1

Merci pour votre attention

HALL OF FAME Guillaume Jourjon 2008 Dino Matin Lopez Pacheco (post-doc) 2009 Pierre-Ugo Tournoux 2012 R´ emi Diana 2012 Nicolas Kuhn 2013 Golam Sarwar 2014 Guillaume Smith 2014 Victor Alejandro Ramiro Cid 2015 Si Quoc Viet Trang 2015 Guilherm Baudic 2016 Ana¨ ıs Finzi 2018 Jonathan Detchart 2018 Bastien Tauran 2018

ISAE-SUPAERO 2018 47 / 1