SlideShare a Scribd company logo
1 of 28
Performance assessment of the MASQUE extension for
proxying scenarios in the QUIC transport protocol
Anno accademico 2022-2023
DIPARTIMENTO DI INGEGNERIA E ARCHITETTURA
Corso di Laurea in Ingegneria Elettronica e Informatica
Curriculum Reti & IoT
Laureando
Alessandro Nuzzi
Relatori
Prof. Alberto Bartoli
Prof. Martino Trevisan
Context
• The web relies on many protocols to enable communication and
information sharing across the Internet
• HTTP and TCP are the predominant protocols driving today's web
infrastructure
• The new QUIC (Quick UDP Internet Connections) transport protocol is
on the rise across the world for many applications
QUIC vs TCP
• Reduced connection establishment
latency
• Connection migration support
• Header encryption
• No head-of-line blocking
QUIC: Usage
• More than 40% of traffic for big
applications requiring high efficiency
• Facebook, Netflix, YouTube, Instagram
• High traffic around the world
• 46% in Latin America, 42% in Europe and
32% in the United States
• Widely supported by providers
• Akamai, Cloudflare, Fastly
Proxies
• Act as intermediaries between clients and servers
• Provide several features
• Network optimization
• Privacy enhancement
• Content filtering
Problem introduction
• QUIC encrypts almost all of its packet header fields
• Difficult for proxies to inspect traffic and apply policy rules
• Proxies might treat QUIC traffic as malicious or unknown
• Traffic delayed or dropped
• Need for new proxying technologies that allow HTTP to create tunnels
for proxying any non-TCP-based protocols, such as QUIC
MASQUE
Multiplexed Application Substrate over QUIC Encryption
• IETF working group formed in June 2020
• Supports proxying UDP and IP over HTTP
• Using QUIC DATAGRAMS with HTTP/3 (RFC 9297)
• CONNECT-UDP (RFC 9298)
• Provides privacy guarantees
• Hiding client IP address from the target server
• Obfuscating traffic destination from client network provider
• Can perform network translation or DNS resolution
CONNECT-UDP
• Uses new DATAGRAM frame
• Unreliable
• Congestion-controlled
• ACK-eliciting, not retransmitted
• Creates and connects a UDP
socket to a target server
• Creates an end-to-end flow of
unreliable messages between
client and server
Thesis objectives
• Evaluate the performance of the MASQUE proposal in proxying
scenarios
• Several network conditions
• Comparison with traditional HTTP/TCP proxies
• Scenarios in which MASQUE usage could be beneficial
Tools used
Hard to find stable implementations of MASQUE and QUIC
• Cloudflare QUIC server
• Google QUIC client
• Google MASQUE client and MASQUE proxy
• curl command-line tool
• Squid proxy
• Twisted HTTPS server
Testing environment
• Docker-based emulation
• Client, server and proxy containers
• Traffic control and network conditions with tc
• Additional delay, bandwidth limit, packet loss
• Automation with Bash scripts
• Execution of repetitive or complex operations
Methodology
• Client requesting a constant-sized file via HTTP GET
• File requested to target server, either through proxy or not
• Transfer time measured
• Different file sizes
• Four categories
• Without proxy: TCP with TLS and QUIC
• With proxy: TCP with TLS and MASQUE
• Simulated network conditions
• Bandwidth limit, additional delay, packet loss
Network conditions
• Additional delay
• Results in additional RTT
• Without proxy: doubled delay applied to client interface
• With proxy: single delays applied to client and server interfaces
• Packet loss
• Without proxy: loss divided by two on all interfaces
• With proxy: loss divided by four on all interfaces
• Bandwidth limit
• Limit applied to client and server interfaces
Experiments
• A full test is made of four experiments, one for each category
• MASQUE, TCP+TLS with proxy, QUIC, TCP+TLS without proxy
• Each experiment is made of 21 measurements
• Delay increased in every new measurement, from 0ms to 200ms
• Each measurement is made of 100 iterations
• Each iteration is a file request
• Fixed network conditions: delay, bandwidth, packet loss
• Data summary based on the iterations measured times
Data
• Collected data for each request
• Real time: elapsed time from client command start to finish
• CPU time: processing time used by client command on user and kernel mode
• Data summary for each measurement
• Collected real times used to calculate measurement summary
• Mean, standard deviation, median, quartiles...
• In case of error, time is discarded and error is logged
Experimental campaigns
• Three scenarios
• Low bandwidth: 10Mbps, no packet loss, variable delay
• Medium bandwidth: 100Mbps, no packet loss, variable delay
• High bandwidth: 1Gbps, no packet loss, variable delay
• In each scenario, 1MB and 10MB download
• Other scenarios
• Medium bandwidth, 1MB file download, 10ms delay, 0%, 1%, 2%, 5% packet loss
Low bandwidth
1MB file
• TCP+TLS with proxy has best
performance
• Gap between QUIC, MASQUE
and TCP+TLS at the beginning
• QUIC similar to TCP+TLS
without proxy
• QUIC has worst performance
between 140ms and 170ms
• QUIC outperforms MASQUE and
TCP+TLS without proxy after
• Internal tuning mechanisms
Medium bandwidth
1MB file
• Same considerations about
TCP+TLS without proxy and
performance gap
• QUIC performs better than
MASQUE, except at 30ms
• QUIC outperforms TCP+TLS
without proxy starting from
80ms, MASQUE from 190ms
• QUIC and MASQUE less
variable than TCP+TLS
High bandwidth
1MB file
• Similar as previous scenario
• QUIC and MASQUE start
performing better than
TCP+TLS without proxy earlier,
at 40ms and 140ms
All bandwidth
10MB file
• Same considerations about
TCP+TLS without proxy and
performance gap
• MASQUE and QUIC never reach
TCP+TLS
• MASQUE perform worse than
QUIC before a certain time,
better after
Medium bandwidth,
1MB, 10ms, loss
• Measured time of each box
increases as loss increases
• MASQUE performs the worst
• QUIC keeps its median below
the TCP+TLS counterpart when
the loss is the highest
CPU time VS Real
time
• QUIC has equal CPU time and
Real time
• The process actively executes
instructions on the CPU
• TCP+TLS cases have constant
CPU time
• Waiting time is not considered in
the CPU time
• MASQUE has almost constant
CPU time, except at the
beginning
Lines overlap!
Results
• TCP+TLS with proxy has the best performance
• In presence of a proxy, TCP+TLS uses two end-to-end independent
connections
• Retransmissions happen independently in the two connections
• Higher throughput
• In MASQUE, only one connection is established
• End-to-end retransmissions
• Using a proxy does not provide advantages
Results (II)
• Small files: MASQUE performs similarly or worse than QUIC
• Bigger files: MASQUE outperforms QUIC in medium and high bandwidth
and high latency links
• Low bandwidth and high latency: QUIC has very good performance
• Lossy links: QUIC has better performance than equivalent TCP+TLS and
than MASQUE
• QUIC and MASQUE have different retransmission mechanism
Results (III)
• Higher bandwidth is beneficial for QUIC and MASQUE with small files
• The higher the bandwidth, the sooner QUIC and MASQUE outperform TCP+TLS
without proxy
• With little to no added delay, QUIC better than MASQUE and TCP+TLS
better than QUIC
• After this time interval, QUIC and MASQUE have small difference
Limitations and future work
• MASQUE early development stage
• Existing implementations are not yet stable
• Need for analysing new or enhanced MASQUE implementations
• Experiments only with single HTTP requests
• Need to evaluate performance of complex web pages
• And other applications: video streaming, mail, etc.
• Need for finding root causes reason of the performance gaps
• Possibility of using more sophisticated network conditions
• Correlated packet loss
• Unevenly distributed delay
Conclusions
• Overall modest performance cost associated with the employment of MASQUE
in proxying scenarios that already use QUIC
• Adopting MASQUE to achieve its privacy guarantees comes at a reasonable
performance expense
• MASQUE can be a good choice in several contexts
• IoT devices
• Companies handling sensitive data
• Developing protocols with encryption and privacy promises involves a
performance trade-off
Thank you for your
attention

More Related Content

Similar to Presentazione-Prelaurea_Alessandro-Nuzzi.pptx

Similar to Presentazione-Prelaurea_Alessandro-Nuzzi.pptx (20)

.NET Conf 2022 - Networking in .NET 7
.NET Conf 2022 - Networking in .NET 7.NET Conf 2022 - Networking in .NET 7
.NET Conf 2022 - Networking in .NET 7
 
WebRTC DataChannels demystified
WebRTC DataChannels demystifiedWebRTC DataChannels demystified
WebRTC DataChannels demystified
 
IAB-5039 : MQTT: A Protocol for the Internet of Things (InterConnect 2015)
IAB-5039 : MQTT: A Protocol for the Internet of Things (InterConnect 2015)IAB-5039 : MQTT: A Protocol for the Internet of Things (InterConnect 2015)
IAB-5039 : MQTT: A Protocol for the Internet of Things (InterConnect 2015)
 
A Quick Look at QUIC, presentation for RIPE 85 by Geoff Huston.pdf
A Quick Look at QUIC, presentation for RIPE 85 by Geoff Huston.pdfA Quick Look at QUIC, presentation for RIPE 85 by Geoff Huston.pdf
A Quick Look at QUIC, presentation for RIPE 85 by Geoff Huston.pdf
 
UDP accelerated file transfer - introducing an FTP replacement and its benefits
UDP accelerated file transfer - introducing an FTP replacement and its benefitsUDP accelerated file transfer - introducing an FTP replacement and its benefits
UDP accelerated file transfer - introducing an FTP replacement and its benefits
 
Messaging for IoT
Messaging for IoTMessaging for IoT
Messaging for IoT
 
how_internet_works.ppt
how_internet_works.ppthow_internet_works.ppt
how_internet_works.ppt
 
Chapter02
Chapter02Chapter02
Chapter02
 
CoAP Talk
CoAP TalkCoAP Talk
CoAP Talk
 
Google QUIC
Google QUICGoogle QUIC
Google QUIC
 
network basics
network basicsnetwork basics
network basics
 
Chapter11
Chapter11Chapter11
Chapter11
 
High performance browser networking ch1,2,3
High performance browser networking ch1,2,3High performance browser networking ch1,2,3
High performance browser networking ch1,2,3
 
Network-Internet
Network-InternetNetwork-Internet
Network-Internet
 
Network-Internet.ppt
Network-Internet.pptNetwork-Internet.ppt
Network-Internet.ppt
 
Network-Internet.ppt
Network-Internet.pptNetwork-Internet.ppt
Network-Internet.ppt
 
Network-Internet.ppt
Network-Internet.pptNetwork-Internet.ppt
Network-Internet.ppt
 
Network-Internet.ppt
Network-Internet.pptNetwork-Internet.ppt
Network-Internet.ppt
 
Network-Internet and network topology.ppt
Network-Internet and network topology.pptNetwork-Internet and network topology.ppt
Network-Internet and network topology.ppt
 
Network-Internet.ppt
Network-Internet.pptNetwork-Internet.ppt
Network-Internet.ppt
 

Recently uploaded

Introduction to Robotics in Mechanical Engineering.pptx
Introduction to Robotics in Mechanical Engineering.pptxIntroduction to Robotics in Mechanical Engineering.pptx
Introduction to Robotics in Mechanical Engineering.pptx
hublikarsn
 
scipt v1.pptxcxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx...
scipt v1.pptxcxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx...scipt v1.pptxcxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx...
scipt v1.pptxcxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx...
HenryBriggs2
 
1_Introduction + EAM Vocabulary + how to navigate in EAM.pdf
1_Introduction + EAM Vocabulary + how to navigate in EAM.pdf1_Introduction + EAM Vocabulary + how to navigate in EAM.pdf
1_Introduction + EAM Vocabulary + how to navigate in EAM.pdf
AldoGarca30
 
Hospital management system project report.pdf
Hospital management system project report.pdfHospital management system project report.pdf
Hospital management system project report.pdf
Kamal Acharya
 
INTERRUPT CONTROLLER 8259 MICROPROCESSOR
INTERRUPT CONTROLLER 8259 MICROPROCESSORINTERRUPT CONTROLLER 8259 MICROPROCESSOR
INTERRUPT CONTROLLER 8259 MICROPROCESSOR
TanishkaHira1
 

Recently uploaded (20)

PE 459 LECTURE 2- natural gas basic concepts and properties
PE 459 LECTURE 2- natural gas basic concepts and propertiesPE 459 LECTURE 2- natural gas basic concepts and properties
PE 459 LECTURE 2- natural gas basic concepts and properties
 
Introduction to Serverless with AWS Lambda
Introduction to Serverless with AWS LambdaIntroduction to Serverless with AWS Lambda
Introduction to Serverless with AWS Lambda
 
litvinenko_Henry_Intrusion_Hong-Kong_2024.pdf
litvinenko_Henry_Intrusion_Hong-Kong_2024.pdflitvinenko_Henry_Intrusion_Hong-Kong_2024.pdf
litvinenko_Henry_Intrusion_Hong-Kong_2024.pdf
 
8th International Conference on Soft Computing, Mathematics and Control (SMC ...
8th International Conference on Soft Computing, Mathematics and Control (SMC ...8th International Conference on Soft Computing, Mathematics and Control (SMC ...
8th International Conference on Soft Computing, Mathematics and Control (SMC ...
 
Online food ordering system project report.pdf
Online food ordering system project report.pdfOnline food ordering system project report.pdf
Online food ordering system project report.pdf
 
Introduction to Robotics in Mechanical Engineering.pptx
Introduction to Robotics in Mechanical Engineering.pptxIntroduction to Robotics in Mechanical Engineering.pptx
Introduction to Robotics in Mechanical Engineering.pptx
 
scipt v1.pptxcxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx...
scipt v1.pptxcxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx...scipt v1.pptxcxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx...
scipt v1.pptxcxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx...
 
UNIT 4 PTRP final Convergence in probability.pptx
UNIT 4 PTRP final Convergence in probability.pptxUNIT 4 PTRP final Convergence in probability.pptx
UNIT 4 PTRP final Convergence in probability.pptx
 
1_Introduction + EAM Vocabulary + how to navigate in EAM.pdf
1_Introduction + EAM Vocabulary + how to navigate in EAM.pdf1_Introduction + EAM Vocabulary + how to navigate in EAM.pdf
1_Introduction + EAM Vocabulary + how to navigate in EAM.pdf
 
Hostel management system project report..pdf
Hostel management system project report..pdfHostel management system project report..pdf
Hostel management system project report..pdf
 
8086 Microprocessor Architecture: 16-bit microprocessor
8086 Microprocessor Architecture: 16-bit microprocessor8086 Microprocessor Architecture: 16-bit microprocessor
8086 Microprocessor Architecture: 16-bit microprocessor
 
Hospital management system project report.pdf
Hospital management system project report.pdfHospital management system project report.pdf
Hospital management system project report.pdf
 
INTERRUPT CONTROLLER 8259 MICROPROCESSOR
INTERRUPT CONTROLLER 8259 MICROPROCESSORINTERRUPT CONTROLLER 8259 MICROPROCESSOR
INTERRUPT CONTROLLER 8259 MICROPROCESSOR
 
COST-EFFETIVE and Energy Efficient BUILDINGS ptx
COST-EFFETIVE  and Energy Efficient BUILDINGS ptxCOST-EFFETIVE  and Energy Efficient BUILDINGS ptx
COST-EFFETIVE and Energy Efficient BUILDINGS ptx
 
Convergence of Robotics and Gen AI offers excellent opportunities for Entrepr...
Convergence of Robotics and Gen AI offers excellent opportunities for Entrepr...Convergence of Robotics and Gen AI offers excellent opportunities for Entrepr...
Convergence of Robotics and Gen AI offers excellent opportunities for Entrepr...
 
Compressing and Sparsifying LLM in GenAI Applications
Compressing and Sparsifying LLM in GenAI ApplicationsCompressing and Sparsifying LLM in GenAI Applications
Compressing and Sparsifying LLM in GenAI Applications
 
S1S2 B.Arch MGU - HOA1&2 Module 3 -Temple Architecture of Kerala.pptx
S1S2 B.Arch MGU - HOA1&2 Module 3 -Temple Architecture of Kerala.pptxS1S2 B.Arch MGU - HOA1&2 Module 3 -Temple Architecture of Kerala.pptx
S1S2 B.Arch MGU - HOA1&2 Module 3 -Temple Architecture of Kerala.pptx
 
NO1 Top No1 Amil Baba In Azad Kashmir, Kashmir Black Magic Specialist Expert ...
NO1 Top No1 Amil Baba In Azad Kashmir, Kashmir Black Magic Specialist Expert ...NO1 Top No1 Amil Baba In Azad Kashmir, Kashmir Black Magic Specialist Expert ...
NO1 Top No1 Amil Baba In Azad Kashmir, Kashmir Black Magic Specialist Expert ...
 
Worksharing and 3D Modeling with Revit.pptx
Worksharing and 3D Modeling with Revit.pptxWorksharing and 3D Modeling with Revit.pptx
Worksharing and 3D Modeling with Revit.pptx
 
Unsatisfied Bhabhi ℂall Girls Ahmedabad Book Esha 6378878445 Top Class ℂall G...
Unsatisfied Bhabhi ℂall Girls Ahmedabad Book Esha 6378878445 Top Class ℂall G...Unsatisfied Bhabhi ℂall Girls Ahmedabad Book Esha 6378878445 Top Class ℂall G...
Unsatisfied Bhabhi ℂall Girls Ahmedabad Book Esha 6378878445 Top Class ℂall G...
 

Presentazione-Prelaurea_Alessandro-Nuzzi.pptx

  • 1. Performance assessment of the MASQUE extension for proxying scenarios in the QUIC transport protocol Anno accademico 2022-2023 DIPARTIMENTO DI INGEGNERIA E ARCHITETTURA Corso di Laurea in Ingegneria Elettronica e Informatica Curriculum Reti & IoT Laureando Alessandro Nuzzi Relatori Prof. Alberto Bartoli Prof. Martino Trevisan
  • 2. Context • The web relies on many protocols to enable communication and information sharing across the Internet • HTTP and TCP are the predominant protocols driving today's web infrastructure • The new QUIC (Quick UDP Internet Connections) transport protocol is on the rise across the world for many applications
  • 3. QUIC vs TCP • Reduced connection establishment latency • Connection migration support • Header encryption • No head-of-line blocking
  • 4. QUIC: Usage • More than 40% of traffic for big applications requiring high efficiency • Facebook, Netflix, YouTube, Instagram • High traffic around the world • 46% in Latin America, 42% in Europe and 32% in the United States • Widely supported by providers • Akamai, Cloudflare, Fastly
  • 5. Proxies • Act as intermediaries between clients and servers • Provide several features • Network optimization • Privacy enhancement • Content filtering
  • 6. Problem introduction • QUIC encrypts almost all of its packet header fields • Difficult for proxies to inspect traffic and apply policy rules • Proxies might treat QUIC traffic as malicious or unknown • Traffic delayed or dropped • Need for new proxying technologies that allow HTTP to create tunnels for proxying any non-TCP-based protocols, such as QUIC
  • 7. MASQUE Multiplexed Application Substrate over QUIC Encryption • IETF working group formed in June 2020 • Supports proxying UDP and IP over HTTP • Using QUIC DATAGRAMS with HTTP/3 (RFC 9297) • CONNECT-UDP (RFC 9298) • Provides privacy guarantees • Hiding client IP address from the target server • Obfuscating traffic destination from client network provider • Can perform network translation or DNS resolution
  • 8. CONNECT-UDP • Uses new DATAGRAM frame • Unreliable • Congestion-controlled • ACK-eliciting, not retransmitted • Creates and connects a UDP socket to a target server • Creates an end-to-end flow of unreliable messages between client and server
  • 9. Thesis objectives • Evaluate the performance of the MASQUE proposal in proxying scenarios • Several network conditions • Comparison with traditional HTTP/TCP proxies • Scenarios in which MASQUE usage could be beneficial
  • 10. Tools used Hard to find stable implementations of MASQUE and QUIC • Cloudflare QUIC server • Google QUIC client • Google MASQUE client and MASQUE proxy • curl command-line tool • Squid proxy • Twisted HTTPS server
  • 11. Testing environment • Docker-based emulation • Client, server and proxy containers • Traffic control and network conditions with tc • Additional delay, bandwidth limit, packet loss • Automation with Bash scripts • Execution of repetitive or complex operations
  • 12. Methodology • Client requesting a constant-sized file via HTTP GET • File requested to target server, either through proxy or not • Transfer time measured • Different file sizes • Four categories • Without proxy: TCP with TLS and QUIC • With proxy: TCP with TLS and MASQUE • Simulated network conditions • Bandwidth limit, additional delay, packet loss
  • 13. Network conditions • Additional delay • Results in additional RTT • Without proxy: doubled delay applied to client interface • With proxy: single delays applied to client and server interfaces • Packet loss • Without proxy: loss divided by two on all interfaces • With proxy: loss divided by four on all interfaces • Bandwidth limit • Limit applied to client and server interfaces
  • 14. Experiments • A full test is made of four experiments, one for each category • MASQUE, TCP+TLS with proxy, QUIC, TCP+TLS without proxy • Each experiment is made of 21 measurements • Delay increased in every new measurement, from 0ms to 200ms • Each measurement is made of 100 iterations • Each iteration is a file request • Fixed network conditions: delay, bandwidth, packet loss • Data summary based on the iterations measured times
  • 15. Data • Collected data for each request • Real time: elapsed time from client command start to finish • CPU time: processing time used by client command on user and kernel mode • Data summary for each measurement • Collected real times used to calculate measurement summary • Mean, standard deviation, median, quartiles... • In case of error, time is discarded and error is logged
  • 16. Experimental campaigns • Three scenarios • Low bandwidth: 10Mbps, no packet loss, variable delay • Medium bandwidth: 100Mbps, no packet loss, variable delay • High bandwidth: 1Gbps, no packet loss, variable delay • In each scenario, 1MB and 10MB download • Other scenarios • Medium bandwidth, 1MB file download, 10ms delay, 0%, 1%, 2%, 5% packet loss
  • 17. Low bandwidth 1MB file • TCP+TLS with proxy has best performance • Gap between QUIC, MASQUE and TCP+TLS at the beginning • QUIC similar to TCP+TLS without proxy • QUIC has worst performance between 140ms and 170ms • QUIC outperforms MASQUE and TCP+TLS without proxy after • Internal tuning mechanisms
  • 18. Medium bandwidth 1MB file • Same considerations about TCP+TLS without proxy and performance gap • QUIC performs better than MASQUE, except at 30ms • QUIC outperforms TCP+TLS without proxy starting from 80ms, MASQUE from 190ms • QUIC and MASQUE less variable than TCP+TLS
  • 19. High bandwidth 1MB file • Similar as previous scenario • QUIC and MASQUE start performing better than TCP+TLS without proxy earlier, at 40ms and 140ms
  • 20. All bandwidth 10MB file • Same considerations about TCP+TLS without proxy and performance gap • MASQUE and QUIC never reach TCP+TLS • MASQUE perform worse than QUIC before a certain time, better after
  • 21. Medium bandwidth, 1MB, 10ms, loss • Measured time of each box increases as loss increases • MASQUE performs the worst • QUIC keeps its median below the TCP+TLS counterpart when the loss is the highest
  • 22. CPU time VS Real time • QUIC has equal CPU time and Real time • The process actively executes instructions on the CPU • TCP+TLS cases have constant CPU time • Waiting time is not considered in the CPU time • MASQUE has almost constant CPU time, except at the beginning Lines overlap!
  • 23. Results • TCP+TLS with proxy has the best performance • In presence of a proxy, TCP+TLS uses two end-to-end independent connections • Retransmissions happen independently in the two connections • Higher throughput • In MASQUE, only one connection is established • End-to-end retransmissions • Using a proxy does not provide advantages
  • 24. Results (II) • Small files: MASQUE performs similarly or worse than QUIC • Bigger files: MASQUE outperforms QUIC in medium and high bandwidth and high latency links • Low bandwidth and high latency: QUIC has very good performance • Lossy links: QUIC has better performance than equivalent TCP+TLS and than MASQUE • QUIC and MASQUE have different retransmission mechanism
  • 25. Results (III) • Higher bandwidth is beneficial for QUIC and MASQUE with small files • The higher the bandwidth, the sooner QUIC and MASQUE outperform TCP+TLS without proxy • With little to no added delay, QUIC better than MASQUE and TCP+TLS better than QUIC • After this time interval, QUIC and MASQUE have small difference
  • 26. Limitations and future work • MASQUE early development stage • Existing implementations are not yet stable • Need for analysing new or enhanced MASQUE implementations • Experiments only with single HTTP requests • Need to evaluate performance of complex web pages • And other applications: video streaming, mail, etc. • Need for finding root causes reason of the performance gaps • Possibility of using more sophisticated network conditions • Correlated packet loss • Unevenly distributed delay
  • 27. Conclusions • Overall modest performance cost associated with the employment of MASQUE in proxying scenarios that already use QUIC • Adopting MASQUE to achieve its privacy guarantees comes at a reasonable performance expense • MASQUE can be a good choice in several contexts • IoT devices • Companies handling sensitive data • Developing protocols with encryption and privacy promises involves a performance trade-off
  • 28. Thank you for your attention