Evolving the Internet Through COVID-19 and Beyond
7 July 2020
Co-authored by Jari Arkko, Alissa Cooper, Tommy Pauly, and Colin Perkins
This article was originally published at CircleID
As we approach four months since the WHO declared COVID-19 to be a
pandemic, and with lock-downs and other restrictions continuing in much
of the world, it is worth reflecting on how the Internet has coped with
the changes in its use, and on what lessons we can learn from these for
the future of the network.
The people and companies that build and operate the Internet are always
planning for more growth in Internet traffic. But the levels of growth
seen since the start of the global COVID-19 pandemic are nothing that
anyone had in their plans. Many residential and mobile networks and
Internet Exchange Points reported 20% traffic growth or more in a
matter of weeks as social distancing sent in-person activities online
around the world. While nearly all kinds of traffic, including web
browsing, video streaming, and online gaming have seen significant
increases, real-time voice and video have seen the most staggering
growth: jumps of more than 200% in traffic and daily conferencing
minutes together with 20-fold increases in users of conferencing
platforms.
By and large, the Internet has withstood the brunt of these traffic
surges. While users have experienced brief outages, occasional video
quality reduction, and connectivity problems now and again, on the
whole, the Internet is delivering as the go-to global communications
system allowing people to stay connected, carry on with their daily
lives from their homes, and coordinate responses to the pandemic.
These are impressive results for a technology designed 50 years ago.
But the robustness of the Internet in the face of huge traffic surges
and shifting usage patterns is no accident. It is the result of
continuous improvements in technology, made possible by its underlying
flexible design. Some recent proposals have suggested that the
Internet's architecture and underlying technologies are not fit for
purpose, or will not be able to evolve to accommodate changing usage
patterns in coming years. The Internet's resiliency in the face of
recent traffic surges is just the latest and most obvious illustration
of why such arguments should be viewed skeptically. The Internet has a
unique model of evolution that has served the world well, continues to
accelerate, and is well-positioned to meet the challenges and
opportunities that the future holds.
The Internet is evolving faster today than ever before
As requirements and demands on networks have changed, the Internet's
ability to continually evolve and rise to new challenges has proved to
be one its greatest strengths. While many people still equate the
Internet with "TCP/IP," the Internet has changed radically over the
past several decades. We have seen a huge increase in scale, great
improvements in performance and reliability, and major advances in
security. Most remarkably, this evolution has been almost entirely
seamless — users notice improvement in their online experience without
any sense of the hard work that went into it behind the scenes.
Even from its early days, the Internet's fundamental structure has
evolved. When the mappings of host names to addresses became
impractical to store and use, the Domain Name System (DNS) was created.
Then the original rigid class-based IP address space was made more
flexible with Classless Inter-Domain Routing, which enabled greater
scalability.
The pace at which major new innovations like these are integrated into
the network has accelerated in recent years. For example, five years
ago about 70% of web connections were not secured with encryption. They
were vulnerable to observation by anyone who intercepted them. But a
renewed focus on security and rapid changes that have made it easier to
deploy and manage security certificates have accelerated encryption on
the web to the point where just 20% or so of web connections are
vulnerable today.
Security protocols are also being updated more rapidly to stay ahead of
attacks and vulnerabilities. Transport Layer Security (TLS) is one of
the foremost protocols used to encrypt application data on the
Internet. The latest version, TLS 1.3, can cut connection setup time in
half, and expands the amount of information that is protected during
setup. The protocol was also carefully designed to ensure that it could
be deployed on the Internet in the broadest way possible. After its
finalization in 2018, there was more TLS 1.3 use in its first five
months than TLS 1.2 saw in its first five years. Roughly one third of
all traffic using TLS has upgraded from TLS 1.2 to TLS 1.3 in a period
of 18 months.
Even the basic transport protocols used on the Internet are evolving.
For decades, there has been a desire to add features to transports that
the original TCP protocol could not support: multiplexed streams,
faster setup time, built-in security, greater data efficiency, and the
ability to use multiple paths. QUIC is a new protocol that supports all
of those features, and is carefully designed with deployability and
protocol evolution in mind. Even prior to finalization, initial
versions of QUIC have become predominant in traffic flows from YouTube
and mobile Facebook applications. QUIC is the foundation for the newest
version of HTTP, HTTP/3. This means that faster and more resilient
connections can be provided to existing HTTP applications, while
opening up new capabilities for future development.
These are just a handful of examples. We have seen the Internet evolve
through massive technological shifts, from the rise of cellular data
networks and mobile broadband to the explosion of voice, video, and
gaming online. We have seen the creation of massively distributed
content delivery networks and cloud computing, the integration of
streaming and conversational multimedia on the web platform, and the
connection of billions of constrained "Internet of Things" devices to
the network. And although some Internet systems have been deployed for
decades, the pace of technological advancement on the network continues
to accelerate.
Keys to successful evolution
This evolvability is an inherent feature of the Internet's design, not
a by-product. The key to successfully evolving the Internet has been to
leverage its foundational design principles while incorporating decades
of experience that teach us how to successfully upgrade a network
composed of billions of active nodes all while it is fully operational
— a process more colloquially known as "changing the engines while in
flight."
The Internet was explicitly designed as a general-purpose network. It
is not tailored to a particular application or generation of
technology. This is the very property that has allowed it to work well
as physical networks have evolved from modems to fiber or 5G and adapt
to traffic shifts like the ones caused by the current pandemic.
Optimizations for particular applications have frequently been
contemplated. For example, "next generation networking" efforts
throughout the decades have insisted on the need for built-in,
fine-grained quality-of-service mechanisms in order to support
real-time applications like voice, video, and augmented reality. But in
practice, those applications are flourishing like never before by
capitalizing on the general-purpose Internet, optimizations in
application design, increased bandwidth, and the availability of
different tiers of Internet service.
Modularity goes hand-in-hand with general-purpose design. Internet
networks and applications are built from modular building blocks that
software developers, architects, network operators, and infrastructure
providers combine in numerous different ways to suit their own needs
while interoperating with the rest of the network. This means that when
it comes time to develop a new innovation, there are abundant existing
software stacks, libraries, management tools, and engineering
experiences to leverage directly. Tools like IP, DNS, MPLS, HTTP, RTP,
and TLS have been re-used so many times that their common usage and
extension models and software support are widely understood.
The Internet was also designed for global reach. Endpoints throughout
the Internet are capable of reaching each other using common systems of
addresses and names even if their local networks use vastly different
underlying technologies. Introducing new or replacement addressing or
naming schemes intended for global reach therefore requires either
complex systems of gateways to bridge between existing
Internet-connected systems and new networks or an incentive structure
that would cause the majority of existing nodes to abandon the Internet
and join the new network. Neither of these offers an obvious path to
seamless global interoperability. And gateways would likely constrain
future evolution across all layers of the stack.
We have seen struggles over incentives play out with the decades-long
advance of IPv6 deployment, as well as with other protocol upgrade
designs like DNSSEC and BGPsec. Experience with the development and
deployment of these protocols has shown that baking deployment
incentives into the design of a protocol itself is key to widespread
deployment. Understanding which actors in the industry will be
motivated to invest in upgrades and having the protocol design place
the onus on those actors is critical.
The TLS 1.3 and QUIC examples highlighted above took these lessons to
heart. Both protocols bind security upgrades together with performance
improvements, knowing that Internet businesses will invest to achieve
better performance and thereby improve security in the process. QUIC
likewise allows application developers to deploy without having to rely
on operating system vendors or network operators to apply updates,
easing the path to widespread adoption.
Testing network innovations at scale in parallel with designing network
protocols is also crucial. In the last five years, every major new
Internet protocol design effort has been accompanied by the parallel
development of multiple (sometimes a dozen or more) independent
implementations. This creates extremely valuable feedback loops between
the people designing the protocols and the people writing the code, so
that bugs or issues found in implementations can lead to quick changes
to the design, and design changes can be quickly reflected in
implementations. Tests of early implementations at scale help to
motivate involvement in the design process from a broader range of
application developers, network operators, equipment vendors, and
users.
Finally, the Internet uses a collaborative model of development:
designs are the product of a community working together. This ensures
that protocols serve the multitude of Internet-connected entities,
rather than serving a limited set of interests. This model also helps
to validate that updates to the network can and will find their way
into production-quality systems. Many academic research efforts focused
on future Internet designs have missed this component, causing their
efforts to falter even with brilliant ideas.
Challenges and opportunities
The Internet faces many technical challenges today and new challenges
will continue to arise in the future. At the same time, technological
advancements and societal changes create opportunities for the Internet
to continue to evolve and meet new needs as they arise.
Security is a multifaceted challenge that has been and continues to be
a major area of evolution. Encryption of many kinds of Internet traffic
is at an all-time high, yet work remains to mitigate unintentional data
leaks and operationalize encryption in core infrastructure services,
such as DNS. Strong protections and mitigations are needed against
threats as diverse as commercial surveillance, denial-of-service
attacks, and malware — all in the context of an Internet that is
increasingly connecting devices that are constrained by computational
power and limited software development budgets. These aspects must be
addressed without intensifying industry consolidation. All of these
challenges are increasingly the focus of Internet protocol designers.
Performance will also continue to require improvements in order to meet
increasing demands. Protocol designers are only just beginning to sort
through how they might leverage the performance gains from QUIC and
HTTP/3 for numerous future applications. Scaling up deployment of
mechanisms such as active queue management (AQM) for reducing latency,
increasing throughput, and managing traffic queues will be needed to
handle an ever-changing mix of traffic flows. Innovative approaches
such as information-centric networking, network coding, moving
computation into the network, establishing common architectures between
data centers and edge networks, decentralised infrastructure and
integration of quantum technology are the focus of ongoing exploration
to respond to current and future performance requirements.
The kinds of networks and devices that benefit from global Internet
connectivity or local IP networking (or both) will continue to
diversify, ranging from industrial to vehicular to agricultural
settings and beyond. Technologies such as deterministic networking,
which seeks to provide latency, loss, and reliability guarantees, and
new protocols explicitly designed to account for intermittent
connectivity and high mobility will all be in the mix as information
technology and operational technology continue to converge.
Conclusion
The Internet of 2020 is vastly different from the one where TCP/IP
originated, even though variants of those original protocols continue
to provide global connectivity. The combination of a general-purpose
design, modularity, global reach, and a collaborative engineering model
with lessons learned about incentives, implementation, and testing at
scale have produced the Internet's winning formula for evolution.
The Internet's unique approach to evolution positions it well as a
technology to meet new challenges and seize new opportunities. The
central role of the Internet in society, only underlined by the
COVID-19 crisis, continues to increase. We hope to never again
experience a crisis that causes such disruption and suffering
throughout the world, but we are optimistic that, crisis or not, the
Internet will continue to evolve to better serve the needs of its
users.
About the authors
Author affiliations are provided for identification and do not imply
organizational endorsement.
Jari Arkko is a member of the
Internet Architecture Board and a Senior Expert with Ericsson Research.
Alissa Cooper is the Internet
Engineering Task Force Chair and a Fellow at Cisco Systems.
Tommy
Pauly is a member of the Internet Architecture Board and an
Engineer at Apple.
Colin Perkins is the Internet
Research Task Force Chair and an Associate Professor at the University
of Glasgow.