Workshop Report

Reducing Internet Latency

Matt Ford, Gorry Fairhurst

Research output: Other contribution

Abstract

This paper reports on a workshop convened to develop an action plan to reduce Internet latency.

Internet latency has become a focus of attention at the leading edge of the industry as the desire to make Internet applications more responsive outgrows the ability of increased bandwidth to address this problem. There are fundamental limits to the extent to which latency can be reduced, but there is considerable capac- ity for improvement throughout the system, making Internet la- tency a multifaceted challenge. Perhaps the greatest challenge of all is to re-educate the mainstream of the industry to understand that bandwidth is not the panacea, and other optimizations, such as reducing packet loss, are at odds with latency reduction.

For Internet applications, reducing the latency impact of shar- ing the communications medium with other users and applica- tions is key. Current Internet network devices were often designed with a belief that additional buffering would reduce packet loss. In practice, this additional buffering leads to intermittently exces- sive latency and even greater packet loss under saturating load. For this reason, getting smarter queue management techniques more widely deployed is a high priority. We can reduce these intermittent increases in delay, sometimes by up to two orders of magnitude, by shifting the focus from packet loss avoidance to delay avoidance using technology that we already have developed, tested, implemented and deployed today.

There is also plenty of scope for removing other major sources of delay. For instance, connecting to a website could be completed in one roundtrip (the time it takes for packets to travel from source to destination and back again) rather than three or four, by folding two or three rounds of flow and security set-up into the first data exchange, without compromising security or efficiency.
Motivating the industry to deploy these advances needs to be aided by the availability of mass-market latency testing tools that could give consumers the information they need to gravitate to- wards low latency services, providers and products. There is no single network latency metric but several alternatives have been identified that compactly express aggregate delay (e.g. as rela- tionships or a constellation), and tools that make use of these will give greater insight into the impact of changes and the diver- sity of Internet connections around the world.
In many developing countries (and in rural regions of developed countries), aside from Internet access itself, there are significant structural issues, such as trombone routes through the developed world and a lack of content distribution networks (CDNs), that need to be addressed with more urgency than Active Queue Man- agement (AQM) deployment, but the ‘blank slate’ of new deploy- ments provides an opportunity to consider latency now. More widespread use of Internet exchange points for hosting local con- tent and fostering local interconnections is key to addressing some of these structural challenges.
Original languageEnglish
Media of outputOn-Line Report
PublisherACM Press
Publication statusPublished - 2014

Fingerprint

Internet
Packet loss
Bandwidth
Slate
Industry
Electronic data interchange
Electric power distribution
Developing countries
Websites
Availability
Communication
Testing

Cite this

Workshop Report : Reducing Internet Latency. / Ford, Matt; Fairhurst, Gorry.

ACM Press. 2014, .

Research output: Other contribution

@misc{f4cbbc971c9b456abbf941044cf142e7,
title = "Workshop Report: Reducing Internet Latency",
abstract = "This paper reports on a workshop convened to develop an action plan to reduce Internet latency.Internet latency has become a focus of attention at the leading edge of the industry as the desire to make Internet applications more responsive outgrows the ability of increased bandwidth to address this problem. There are fundamental limits to the extent to which latency can be reduced, but there is considerable capac- ity for improvement throughout the system, making Internet la- tency a multifaceted challenge. Perhaps the greatest challenge of all is to re-educate the mainstream of the industry to understand that bandwidth is not the panacea, and other optimizations, such as reducing packet loss, are at odds with latency reduction.For Internet applications, reducing the latency impact of shar- ing the communications medium with other users and applica- tions is key. Current Internet network devices were often designed with a belief that additional buffering would reduce packet loss. In practice, this additional buffering leads to intermittently exces- sive latency and even greater packet loss under saturating load. For this reason, getting smarter queue management techniques more widely deployed is a high priority. We can reduce these intermittent increases in delay, sometimes by up to two orders of magnitude, by shifting the focus from packet loss avoidance to delay avoidance using technology that we already have developed, tested, implemented and deployed today.There is also plenty of scope for removing other major sources of delay. For instance, connecting to a website could be completed in one roundtrip (the time it takes for packets to travel from source to destination and back again) rather than three or four, by folding two or three rounds of flow and security set-up into the first data exchange, without compromising security or efficiency.Motivating the industry to deploy these advances needs to be aided by the availability of mass-market latency testing tools that could give consumers the information they need to gravitate to- wards low latency services, providers and products. There is no single network latency metric but several alternatives have been identified that compactly express aggregate delay (e.g. as rela- tionships or a constellation), and tools that make use of these will give greater insight into the impact of changes and the diver- sity of Internet connections around the world.In many developing countries (and in rural regions of developed countries), aside from Internet access itself, there are significant structural issues, such as trombone routes through the developed world and a lack of content distribution networks (CDNs), that need to be addressed with more urgency than Active Queue Man- agement (AQM) deployment, but the ‘blank slate’ of new deploy- ments provides an opportunity to consider latency now. More widespread use of Internet exchange points for hosting local con- tent and fostering local interconnections is key to addressing some of these structural challenges.",
author = "Matt Ford and Gorry Fairhurst",
note = "Note many contributing authors listed in the report",
year = "2014",
language = "English",
publisher = "ACM Press",
type = "Other",

}

TY - GEN

T1 - Workshop Report

T2 - Reducing Internet Latency

AU - Ford, Matt

AU - Fairhurst, Gorry

N1 - Note many contributing authors listed in the report

PY - 2014

Y1 - 2014

N2 - This paper reports on a workshop convened to develop an action plan to reduce Internet latency.Internet latency has become a focus of attention at the leading edge of the industry as the desire to make Internet applications more responsive outgrows the ability of increased bandwidth to address this problem. There are fundamental limits to the extent to which latency can be reduced, but there is considerable capac- ity for improvement throughout the system, making Internet la- tency a multifaceted challenge. Perhaps the greatest challenge of all is to re-educate the mainstream of the industry to understand that bandwidth is not the panacea, and other optimizations, such as reducing packet loss, are at odds with latency reduction.For Internet applications, reducing the latency impact of shar- ing the communications medium with other users and applica- tions is key. Current Internet network devices were often designed with a belief that additional buffering would reduce packet loss. In practice, this additional buffering leads to intermittently exces- sive latency and even greater packet loss under saturating load. For this reason, getting smarter queue management techniques more widely deployed is a high priority. We can reduce these intermittent increases in delay, sometimes by up to two orders of magnitude, by shifting the focus from packet loss avoidance to delay avoidance using technology that we already have developed, tested, implemented and deployed today.There is also plenty of scope for removing other major sources of delay. For instance, connecting to a website could be completed in one roundtrip (the time it takes for packets to travel from source to destination and back again) rather than three or four, by folding two or three rounds of flow and security set-up into the first data exchange, without compromising security or efficiency.Motivating the industry to deploy these advances needs to be aided by the availability of mass-market latency testing tools that could give consumers the information they need to gravitate to- wards low latency services, providers and products. There is no single network latency metric but several alternatives have been identified that compactly express aggregate delay (e.g. as rela- tionships or a constellation), and tools that make use of these will give greater insight into the impact of changes and the diver- sity of Internet connections around the world.In many developing countries (and in rural regions of developed countries), aside from Internet access itself, there are significant structural issues, such as trombone routes through the developed world and a lack of content distribution networks (CDNs), that need to be addressed with more urgency than Active Queue Man- agement (AQM) deployment, but the ‘blank slate’ of new deploy- ments provides an opportunity to consider latency now. More widespread use of Internet exchange points for hosting local con- tent and fostering local interconnections is key to addressing some of these structural challenges.

AB - This paper reports on a workshop convened to develop an action plan to reduce Internet latency.Internet latency has become a focus of attention at the leading edge of the industry as the desire to make Internet applications more responsive outgrows the ability of increased bandwidth to address this problem. There are fundamental limits to the extent to which latency can be reduced, but there is considerable capac- ity for improvement throughout the system, making Internet la- tency a multifaceted challenge. Perhaps the greatest challenge of all is to re-educate the mainstream of the industry to understand that bandwidth is not the panacea, and other optimizations, such as reducing packet loss, are at odds with latency reduction.For Internet applications, reducing the latency impact of shar- ing the communications medium with other users and applica- tions is key. Current Internet network devices were often designed with a belief that additional buffering would reduce packet loss. In practice, this additional buffering leads to intermittently exces- sive latency and even greater packet loss under saturating load. For this reason, getting smarter queue management techniques more widely deployed is a high priority. We can reduce these intermittent increases in delay, sometimes by up to two orders of magnitude, by shifting the focus from packet loss avoidance to delay avoidance using technology that we already have developed, tested, implemented and deployed today.There is also plenty of scope for removing other major sources of delay. For instance, connecting to a website could be completed in one roundtrip (the time it takes for packets to travel from source to destination and back again) rather than three or four, by folding two or three rounds of flow and security set-up into the first data exchange, without compromising security or efficiency.Motivating the industry to deploy these advances needs to be aided by the availability of mass-market latency testing tools that could give consumers the information they need to gravitate to- wards low latency services, providers and products. There is no single network latency metric but several alternatives have been identified that compactly express aggregate delay (e.g. as rela- tionships or a constellation), and tools that make use of these will give greater insight into the impact of changes and the diver- sity of Internet connections around the world.In many developing countries (and in rural regions of developed countries), aside from Internet access itself, there are significant structural issues, such as trombone routes through the developed world and a lack of content distribution networks (CDNs), that need to be addressed with more urgency than Active Queue Man- agement (AQM) deployment, but the ‘blank slate’ of new deploy- ments provides an opportunity to consider latency now. More widespread use of Internet exchange points for hosting local con- tent and fostering local interconnections is key to addressing some of these structural challenges.

M3 - Other contribution

PB - ACM Press

ER -