Internet-Draft NOISV October 2024
Deen & Mishra Expires 13 April 2025 [Page]
Workgroup:
Media OPerationS
Internet-Draft:
draft-deen-mops-network-overlay-impacts-02
Published:
Intended Status:
Informational
Expires:
Authors:
G. Deen
Comcast-NBCUniversal
S. Mishra
Verizon

Network Overlay Impacts to Streaming Video

Abstract

This document examines the operational impacts to streaming video applications caused by changes to network policies by network overlays. The network policy changes include IP address assignment, transport protocols, routing, DNS resolver which in turn affect a variety of important content delivery aspects such as latency, CDN cache selection, delivery path choices, traffic classification and content access controls.

About This Document

This note is to be removed before publishing as an RFC.

The latest revision of this draft can be found at https://gitnnelg.github.io/NetworkOverlays/draft-deen-mops-network-overlay-impacts.html. Status information for this document may be found at https://datatracker.ietf.org/doc/draft-deen-mops-network-overlay-impacts/.

Discussion of this document takes place on the Media OPerationS Working Group mailing list (mailto:mops@ietf.org), which is archived at https://mailarchive.ietf.org/arch/browse/mops/. Subscribe at https://www.ietf.org/mailman/listinfo/mops/.

Source for this draft and an issue tracker can be found at https://github.com/gitnnelg/NetworkOverlays.

Status of This Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."

This Internet-Draft will expire on 13 April 2025.

Table of Contents

1. Introduction

The past decade of Internet evolution has included two significant trends, the global growth of video streaming and active passionate work within the IETF on enhancing Internet user privacy.

The work on these initiatives has largely occured independently of one another, though there are a few individuals and companies that are involved with both efforts.

The arrival of the newly developed privacy enhancements in consumer products and their subsequent use by streaming video viewers has brought the results of the two efforts into contact and a number of friction points are now being encountered which are having impacts to the viewers, support engineers and operational aspects of video streaming platforms.

To be clear, this document is not proposing or advocating rolling back any of the privacy enhancements for the viewers. Instead the authors hope to help describe the problem space and educate the IETF and others on the practical operational impacts of these enhancements and to eventually develop approaches that can help mitigate such impacts.

The authors also readily acknowledge the many challenges and difficulties in improving Internet privacy in an area as complex as the Internet while also maintaining compatibiltiy with the wildly varied applications and uses of the Internet on which users rely upon daily in their lives. This is hard stuff and it's very natural for there to be operational considerations that must be understood and folded back into architectural designs and consumer products.

The motivation in developing this document is to provide a meaningful and helpful feedback from the streaming application and streaming platform operational perspective to help the enhanced privacy architecture work being done at the IETF.

2. Internet Privacy Enhancements

Enhancing the Internet's privacy is a difficult challenge, given the complexity of the Internet itself. It's common for solutions that address one issue to inadvertently create new problems elsewhere. That's not a reason to stop trying, but it is important to understand the consequences of changes and to find ways to manage or mitigate such impacts, ideally without weakening or rolling back the enhancements.

A popular design choice in privacy enhancements at the IETF has been the encapsulation of data inside encrypted connections along with other network policy changes to introduce changes which make observing and tracing data difficult to do and difficult to associate to any particular user.

[RFC7258] from the IAB examines various pervasive montoring approaches while [RFC7624] discusses responses that enhance privacy. [RFC9000] itself is an excellent example of the applied design approaches and introduces the QUIC transport protocol that is always encrypted.

2.1. Network Overlays

Along with the use of encrypted connections another popular approach is to additionally create alternative routes and tunnels for connections which bypass the routing and other policy decisions of the ISP access network and of the public open Internet. These alternative network policy choices have the effect of creating a Network Overlay that operates on top of and over the device's Access Network and the Open Internet, but follows an independent set of policies chosen by the Network Overlay.

 R  = router
                  <--- non-overlay traffic path --->
 device -- R ---- R ------------- R ------------- R ---- R -- dest-node
            \                                           /
             \                                         /
              \                                       /
               R -- R -- ingest -- egress -- R ------+
                     <--- overlay traffic path --->

 Figure 1:  Network Overlay routing select traffic via an alternate path

2.1.1. Partitioning

Network Overlay policy changes includes an alternate routing policy since a fundamental aspect of this design is the tunneling of connections through alternate paths to enhance privacy. The reasons for this approach are discussed in the IAB document Partitioning as an Architecture for Privacy.

2.1.2. Policy Changes

Beyond alternate routing policies, network overlays often also make changes to the Source IP Address assignment, and/or selection of the DNS Resolver and/or including protocol conversions/translations such as HTTP2->HTTP3 and HTTP2->HTTPS2+TLS, and can include IP layer changes such as IPv4->IPv6 or IPv6->IPv6 conversions.

2.1.3. MASQUE

Protocols such as MASQUE [RFC9484] and services built on it such as Apple's iCloud Private Relay are examples of Privacy Enhancing Network Overlays that involve making a number of network policy changes from the open Internet for the connections passed through them.

2.2. Making It Easy (for Users) by Working Under the Covers

Historically, incorporating privacy features into consumer-facing products has been complex. This challenge arises from the need to address a wide range of use cases while also offering users easy access to advanced privacy frameworks and taxonomies. Many attempts have been made and very few have achieved finding success with end users.

Perhaps learning from the lessons of offering too many options, the recent trend in privacy enhancements has steered torward either a very simple "Privacy On or Off" switch or in other cases automatically enabling or "upgrading" to enhance privacy. Apple's iCloud Private Relay can be easily turned on with a single settings switch, while privacy features such as Encrypted DNS over HTTP and upgrade from HTTP to HTTPS connections have had a number of deployments that automatically enable them for users when possible.

Keeping with the motto of "Keep It Simple", users are generally not provided with granular Network Overlay controls permitting the user to select what applications, or what network connections the Network Overlay's policies can apply to.

Adhereing with the "Keep It Simple" approach the application itself has very little connection to privacy enhancing Network Overlays. Applications generally do not have a means to detect when networking policy changes are active. Applications generally do not have a means to access policy change settings or to interact to change them.

3. Streaming Video

Streaming Video, while just one of the many different Internet applications does standout from other uses in a number of significant ways that perhaps merit some amount of special consideration in understanding and addressing the impacts caused by particular privacy enhancing design and service offering choices.

Firstly, Streaming video operates at a hard to imagine scale - streaming video is served globally to more than 2 billion user daily currently and continuing to grow in leaps and bounds.

Secondly, the content types delivered through streaming has evolved from the pre-recorded low-resolution, low-bit rate, latency tolerant video-on-demand movies, live or pre-recorded TV shows, and user generated videos delivered by pioneering streaming platforms to now including low-latency 4K and 8K live sports events, while also evolving the pre-recorded content with high-bit rate such as 4K and 8K cinema quality and High Dynamic Range (HDR) lighting.

Finally, the expectations of streaming video viewers have significantly evolved from the days of settling for being able to watch a movie in a PC browser. Viewers expect to watch on any device type they want ranging from low-end-streaming sticks that plug into a USB port, to 4K and HDR capable laptops, 4K and 8K HDR TV screens, gaming consoles, smart phones and many more choices. Viewers also expect to have the same great viewing experience while at home connected via high-speed wired Internet, high-speed WiFi, or mobile cellular 5G and even satellite Internet connections.

To meet the growth to billions of users, the growth in content type, quality and speed expectations, and on-any-device anywhere that I am over any-network-connection expectations of users the Streaming Video technology infrastructure has had to itself evolve significantly. This video streaming evolution work is being done in the IETF and in the Streaming Video Technology Alliance (SVTA), and in a number of other technical and industry groups.

It's hard to overstate just how much the growth of streaming video has contributed to the growth of the Internet. Internet connections of multiples of hundred megabits and gigabits speeds today are because of the needs of video streaming, the ongoing work on low-latency networking and ultra-low-latency video delivery are both driven by the use of streaming video.

3.1. Advances in Streaming Video Architecture

Internet streaming has greatly matured and diversified from its early days of viewers watching pre-recorded 320x240, 640x480 standard definiton 480p movies to wired PCs connected to the Internet via high-latency, low-bandwidth DSL as early DOCSIS modems.

Streaming has grown to the extent that it has become a daily go-to video source world wide for billions of viewers and has expanded from pre-recorded movies to encompass every type of video content imaginable. This growth to billions of viewers and the addition of low latency sensistive content and new connectivity options like WiFi, Cellular and Satellite in addition to high-speed DOCSIS and fiber is the world streaming platforms now provide service in.

With the large user base and its usage, the Streaming platforms also have significant technical challenges to meet viewer expectations:

  • (1) Delivery scales that commonly range from hundreds of thousands to many millions of viewers simultaneously, with billions of views globally daily;

  • (2) Low latency demands from live sports, live events and live streamed content;

  • (3) content resolutions and corresponding formats which have jumped from the days of SD-480p to 4K (3840x2160) and 8K (7680x4320) along with bit rates which can had data needs of 10-24+ Mbps for 4K with 8K demanding 40 Mbps under extreme compression and 150-300 Mbps for high quality such as cinema;

  • (4) devices with very diverse capabilities low-cost streaming sticks, to Smart TVs, tablets, phones, and game consoles

  • (5) broad range of connectivity choices including WiFi, Gig speed-low latency DOCSIS, Fiber, satellite, and 5G cellular networks;

  • (6) application transport protocols including MPEG DASH, HLS, HTTP2/TCP, HTTP3/QUIC, WebRTC, Media over QUIC (MoQ) and specialty application transports such as SRT, HESP etc.

To meet these challenges streaming platforms have significantly invested in developing delivery architectures that are built with detailed understandings of each element in the content delivery pathway starting from the content capture all the way through to the screen of the viewer.

Streaming applications are part of an end-to-end architecture that is optimized around achieving the best experience including low latency video delivery to viewing devices. The open Internet can be unpredictable with temporary issues like packet loss, congestion and other conditions. However, streaming architecture is desiged to handle these momentary problems as effectively as possible often through use of dynamic adaptive approaches designed into streaming protocols and platform components.

4. Emerging Operational Issues with Network Overlay Policy Changes

Streaming video applications and the streaming platforms delivering content are starting to encounter various operational challenges related to Network Overlays. Typically the primary problems are encountered when the network overlay has made policy changes that are either unexpected, are difficult or impossible for the streaming platform to detect, or the changes are inconsitently applied.

There are a variety of impacts but a few common classes of issues have been observed:

4.1. Policy Changes Hidden from Applications

One of the central recurring issues with streaming applications running on devices or networks with changed policies due to network overlays is that the changes are often hidden from the applications.

Applications often find it difficult or even impossible to detect when network policy changes will be active and what they are changing. For example, a device may have a desingated default DNS resolver for the device, but may have a different resolver selected depending on how the streaming application queries the DNS.

Likewise, a streaming application might find that one application transport protocol such as HTTP queries will have one set of routing policies applied to it but a different appliction transport like HTTPS may have a different set of routing policies applied.

Streaming applications that cannot determine the exact behavior to be expected can prevent the streaming application from making good content source decisions and can prevent appllications from being able to provide reliable feedback and logs when problems are encountered.

4.2. Routing Changes

Routing changes which cause connections between video applications and the infrastructure servcices they use can create a large number of problems.

4.2.1. End to End Problem Discovery

A common issue in video delivery is locating where along the delivery path the video transport is encountering problems. Often such problems are more complex than the connection not working at but instead involve identifying bottleneck, lost packets, congestion issues. When the routing changes from what is expected or visible to support tools it becomes an operational trouble spot for users and platform suport to location and determine the source of the problems.

4.2.2. CDN Edge Cache Selection due to Routing

A significant, and often overlooked problm is the addition of network latency compared to edge CDN caches or access network peering connections. Routing changes which cause bypassing edge CDN caches and instead choosing less optimal caches

 R  = router
           <--- non-overlay traffic path --->
 device -- R ---- R ---- Edge CDN Cache
            \
             \
              \
               R --- R -- ingest -- R --- R -- egress -- R ------R ---- Less Optimal CDN Cache
                     <--- overlay traffic path --->

 Figure:  Routing Changes alering CDN Cache selection

4.3. Changed Encryption Policy

Changing the encryption policy applied to video streams either adding where it wasn't orginally used or removing if it was originally specified can cause a wide range of operational problems.

4.3.1. Forced Encryption Upgrade

Changing unencrytped HTTP2 to encrypted HTTP2+TLS connects will prohibit streaming workflows that involve content detection as part of the network delivery. This can result in video traffic not being correctly identified and the incorrect network policies being applied to it. This is particularly problematic in environments using multicast and in mobile environments.

4.3.2. Forced Encryption Downgrade

Equally so, removing of encryption applied to the transport stream by a streaming platform would be significantly problematic as such encryption may be part of a content protection and content integrity protections architecture.

4.4. Changed DNS Resolver Selection

DNS Resolver choice changes resulting in less optimal CDN cache selection or bypassing of CDN load balancing direction

4.5. Changed Source IP Address

Changing the Source IP Address for the application's connections to Streaming Platform Servers resulting in logging, geofencing, and session management problems

4.6. Performance and Problem determination

Network overlays often interfere with the tools used in performance and problem determination. This is due to either the tool and protocols not able to traverse the alternative route tunnel impacting services ability to diagnose connection and performance problems, or the network overlay itself not supporting the tool and not supporting or carrying the tools functions.

4.7. Impact of Changing Network Routing and other Policies

The problem for streaming applications occurs when the underlying network properties and policies change from what is expected by the streaming application. In particular when such changes are either hidden or not visible to the streaming application.

While the open Internet is a dynamic environment, changing of basic network behavior and policies from what is expected as seen from the streaming application, deviate unexpectedly from what the streaming application expects. This behavior disrupts the optimized streaming delivery architecture for the end-user device. Changes to Network Policies such the routing, source IP address assigned to the streaming application traffic, DNS resolver choice etc influences this behavior.

Having a reliable understanding of the delivery path is essential for streaming operators and the introduction of network overlays like those based on technologies such as MASQUE especially when designed to be undetectable by the applications using them has introduced new technical challengess for streaming operators and network operators as well as for their viewers.

The core problem occurs when changes to network policies are made, often without notification or visibilty to applications and without clear methods of probing to determine and test changed behaviors that affect the streaming application's content delivery path resulting in increased latency, changes of IP address for the application as seen by either the application or the streaming service connection, changes to DNS resolvers being queried and the results returned by DNS, and changes to application transports such as adding or removing outer layer encryption are all problems that have been observed in production streaming platforms.

4.7.1. Middleboxes and learning from the past

The IETF has discussed this situation in the past, more than 20 years ago in 2002 Middleboxes: Taxonomy and Issues [RFC3234] was published capturing the issues with Middleboxes in the network and the affects of hidden changes occuring on the network between the sender and receiver.

5. Appendix A: Network Overlays are different than VPNs

While conceptually similar in many ways to VPN (Virtual Private Network) technology, the various network overlay technologies currently being deployed as well as new ones currently being designed by the IETF differ quite siginificanlty from the older VPN approach they are replacing in a number of ways.

It is also worth noting that one reason why the issues discussed in this document have not been concerned with regard to VPNs is that largely VPNs have not been a pervasive way to stream video. First, many VPNs have not had very good or consistent throughput compared to the direct open Internet and so provide a poor viewing experience. Second, many video platforms block or deny service to VPN connections due to the very common use of VPNs to bypass geofiltering restrictions.

Whatever the reason, it's work looking at how VPNs differ from the Network Overlays being discussed herein.

5.1. VPNs typically:

  • (1) VPNs typically are detectable by both the video application and often by the streaming platform.

  • (2) VPNs typically work at the network layer of a device, resulting in a wide-range (if not all) transports

  • and protocols from the device flowing through the VPN

  • (3) VPNs typically provide exception options allowing for exclusion from traversing via the VPN based on

  • various criteria such as application, destination IP address, application protocol etc.

5.1.1. Network Overlays typically:

  • (1) Network Overlays are often undetectable by video applications or by the streaming platform, when in use

  • (2) Network Overlays often only apply to specific application transports such as HTTP2/TCP or HTTP3/QUIC while not applying to HTTP2/TCP+TLS on the same device.

  • (3) Network Overlays often only apply to HTTP connections and do not support ICMP, non-http versions of DNS, NTP etc, and various tools used for network measurement, problem determination, and network management that are not http based.

  • (4) Network Overlays do not expose to applications any means for the application to discover the policy changes the overlay will apply to the applications network connections.

  • (5) Network Overlays do not expose mechanisms or APIs for applications to interact with them such as getting or setting options.

6. Conventions and Definitions

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.

7. Security Considerations

TODO Security

8. IANA Considerations

This document has no IANA actions.

9. Normative References

[RFC2119]
Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, , <https://www.rfc-editor.org/rfc/rfc2119>.
[RFC3234]
Carpenter, B. and S. Brim, "Middleboxes: Taxonomy and Issues", RFC 3234, DOI 10.17487/RFC3234, , <https://www.rfc-editor.org/rfc/rfc3234>.
[RFC7258]
Farrell, S. and H. Tschofenig, "Pervasive Monitoring Is an Attack", BCP 188, RFC 7258, DOI 10.17487/RFC7258, , <https://www.rfc-editor.org/rfc/rfc7258>.
[RFC7624]
Barnes, R., Schneier, B., Jennings, C., Hardie, T., Trammell, B., Huitema, C., and D. Borkmann, "Confidentiality in the Face of Pervasive Surveillance: A Threat Model and Problem Statement", RFC 7624, DOI 10.17487/RFC7624, , <https://www.rfc-editor.org/rfc/rfc7624>.
[RFC8174]
Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, , <https://www.rfc-editor.org/rfc/rfc8174>.
[RFC9000]
Iyengar, J., Ed. and M. Thomson, Ed., "QUIC: A UDP-Based Multiplexed and Secure Transport", RFC 9000, DOI 10.17487/RFC9000, , <https://www.rfc-editor.org/rfc/rfc9000>.
[RFC9484]
Pauly, T., Ed., Schinazi, D., Chernyakhovsky, A., Kühlewind, M., and M. Westerlund, "Proxying IP in HTTP", RFC 9484, DOI 10.17487/RFC9484, , <https://www.rfc-editor.org/rfc/rfc9484>.

Acknowledgments

The authors would like to acknowledge to the contributions from the Streaming Video Technology Alliance (SVTA) based on their work studying the impacts of network overlays on the streaming platforms. In particular, contributions from Brian Paxton have been very helpful.

Authors' Addresses

Glenn Deen
Comcast-NBCUniversal
Sanjay Mishra
Verizon