How QoS differentiation enhances the OTT video streaming experience. Netflix over a QoS enabled

(1)

NSN White paper

Netflix over a QoS enabled LTE network

February 2013

How QoS differentiation enhances the OTT video streaming experience

Netflix over a QoS enabled

LTE network

(2)

Executive summary 3

Why over-the-top video streaming matters to

mobile networks 3

What video users expect from mobile networks 4 How QoS can be used to enhance the OTT video

experience on LTE 5

The impact of QoS on a Netflix session during

congestion 6

Conclusion 11

Abbreviations 12

CONTENTS

(3)

Executive Summary

Cellular networks are more and more impacted by the growing popularity of Over the Top (OTT) video streaming services. The introduction of smartphones and tablets have opened the use of video streaming services which was limited to personal computers and televisions tied to wireline networks in the past.

Consumers expect a high quality experience when watching OTT video streams, whether it be short form YouTube videos or long form movies or television programs from providers such as Netflix. The transport of video streaming in addition to all the other types of traffic such as VoIP, email, web browsing, instant messaging and social media can benefit if Quality of Service (QoS) differentiation is used – in particular during peak usage times. QoS differentiation brings clear improvements to the customer experience.

Nokia Siemens Networks has tested how QoS mechanisms can be used to improve video streaming quality in LTE networks during congestion. The findings from the tests using Netflix point to gains for both maintaining video quality and video streaming performance where re- buffering/stalling of the stream is minimized as compared to best effort transport which creates a sub-standard user experience over a congested LTE network.

QoS mechanisms are an effective tool for operators to provide differentiated delivery of services transporting large payload content. With QoS the existing business models for content delivery can be changed. New content delivery business models can be created which involve the operator as part of the content delivery valuation.

Why over-the-top video streaming matters to mobile networks

OTT video streaming is delivered to the end user over a Internet Service Provider (ISP) or wireless operator by a 3rd party without the operator controlling the service other than to provide transport of the content. The rapidly growing OTT video market is generally considered to include the streaming of content such as movies, television programs, and music videos by content providers such as Netflix, Hulu, Amazon Prime, iTunes, YouTube and others. These providers are globally offering services to both fixed and wireless mobile customers.

Netflix, one of the early OTT video service pioneers, has grown to 32 million video streaming users at the end of 2012. Many OTT video service providers are experiencing additional growth driven by smartphone and tablet users consuming video services over wireless networks. In fact, Network World reported that video accounted for half of total mobile data traffic in 2012

(www.networkworld.com), which increased from 42% in 2011 according to Billing and OSS world (http://www.billingworld.com). By 2017, mobile video will represent 66% of all mobile data traffic according to Cisco. The video share on mobile data traffic will be even higher than in fixed networks (Cisco Virtual Networking Index 2013).

Coincident with the growth in OTT video is a change in its composition. In the past, the bulk of video traffic was composed of short-form, YouTube-ish clips and many may have tolerated low

(4)

resolution video clips that frequently froze, especially when the video was for free. However when customers pay for entire movies or episodes and display them on HD screens, they become very sensitive to anything other than crystal-clear video and audio fidelity. With video services taking an increasing share of the users’ screen time on mobile devices, the importance of high quality is evident.

Operators have the opportunity to differentiate themselves from the competition and change the game in content delivery by enhancing the customer experience of video services. On the other hand scarce network capacity has to be utilized in an economic way.

Figure 1: Global mobile data traffic forecast Source: Cisco Virtual Networking Index 2013

What video users expect from mobile networks

Streaming of video programs such as movies and television over cellular networks in addition to other types of applications like web browsing, social media, email and voice adds a significant load due to the associated size of the transported payload coupled with the need for good throughput speeds (figure 2). Moreover, users expect a high level of service delivery for video without re-buffering, video quality degradation, or slow start times which creates the need for QoS mechanisms to be employed to maintain a high quality of experience (QoE).

Figure 2: Non-HD Video average requirements

Source: 4G Americas / Rysavy Research Mobile Broadband Explosion 2012 Whitepaper 0

2.000.000 4.000.000 6.000.000 8.000.000 10.000.000 12.000.000

2012 2013 2014 2015 2016 2017

M2M File Sharing Data Video

89%

34%

55%

75%

CAGR 2012-2017

Non-HD Video Throughput (Mbps) MB/Hour

Small Screen Video (feature phone)

0.2 90

Medium Screen Video (smartphone)

1.0 450

Large Screen Video (tablet)

2.0 900

(5)

The ability of a cellular network to provide users with a good quality of experience for video varies based on signal strength, interference, and to a large extent the bandwidth available during peak demand periods for the network. To manage congestion, cellular networks need to employ more sophisticated traffic management features to maintain quality for demanding applications like video while still being able to serve other applications which users need.

How QoS can be used to enhance the OTT video experience on LTE

3GPP (Third Generation Partnership Project) designed QoS mechanisms (TS 23.203 Policy and Charging Control Architecture) to allow LTE operators to manage the quality of experience for users based on the application types used on the network.

The nine standardized QoS class identifiers (QCI) address prioritized handling and quality parameters for the many different types of traffic the network has to transport. 3GPP’s QCI concept is illustrated in figure 3.

Figure 3: 3GPP TS 23.203 standardized QoS Class Identifiers (QCI) characteristics

QCI Resource Type

Priority Packet Delay Budget

Packet Error Loss Rate

Example Services

1

GBR

2 100 ms 10^-2 Conversational Voice

2 4 150 ms 10^-3 Conversational Video (Live Streaming)

3 3 50 ms 10^-3 Real Time Gaming

4 5 300 ms 10^-6 Non-Conversational Video (Buffered

Streaming) 5

NON-GBR

1 100 ms 10^-6 IMS Signalling

6 6 300 ms 10^-6

Video (Buffered Streaming)

TCP-based (e.g., www, e-mail, chat, ftp, p2p file sharing, progressive video, etc.)

7 7 100 ms 10^-3 Voice, Video (Live Streaming)

Interactive Gaming

8 8

300 ms 10^-6

Video (Buffered Streaming)

TCP-based (e.g., www, e-mail, chat, ftp, p2p file sharing, progressive video, etc.)

9 9

(6)

QoS Class Identifier (QCI) values define the level of service required by the application.

There are two broad resource type levels: Guaranteed (GBR) and non-Guaranteed (Non-GBR) to support different service types. Services using GBR’s get pre-allocated capacity in the network while non-GBR services are given access as needed.

Netflix as an OTT video streaming service would normally be handled just like all other internet applications over the network with a QCI9 quality treatment which is referred widely as best effort (BE) similarly to other non-operator services such as OTT VoIP. Cellular networks treat best effort applications with the same priority and will schedule the associated transmissions equally.

In normal conditions when there is no congestion, best effort treatment works without noticeable impact on performance.

When congestion in cells develops during peak usage times or when there are multiple

subscribers using bandwidth intense, video streaming applications then best effort is not able to schedule the transmission of data often enough or long enough. The lack of scheduling time from network resources with limited availability results in users experiencing problems with their video stream such as slow start, re-buffering / stalling and degraded video quality.

3GPP does define other QCI levels which could be applied to OTT streaming services: QCI 6 for video with buffered streams for non-GBR services. Video streaming applications are best

supported by a non-GBR resource type due to intermittent usage of network resources versus conversational video which is always sending data.

The impact of QoS on a Netflix session during congestion

To analyze the potential impact of QoS on Netflix sessions in different conditions, a battery of tests was executed in the LTE lab network of Nokia Siemens Networks with the Netflix application running on an Android based Smartphone. The test setup is described in Figure 4. Note that real- world results may be different from lab environment results.

Figure 4: Test setup

(7)

The main test scenario was to use the Netflix application on a smartphone and observe the video performance for any disruptions like re-buffering or for variations in displayed video quality, while the network was being utilized in the background by other users who were loaded incrementally from a UE (User Equipment) simulator as the test progressed.

While the network experienced congestion generated from the UE simulator, Netflix video quality performance and the data rate required to maintain the stream was noted. The Netflix video quality was measured based on the following factors to determine the user experience:

• High – the observed quality of the video being played is excellent and the end user has a very good service experience (no re-buffering)

• Medium – the observed quality of the video being played is slightly degraded but the end user service experience is acceptable and end user would continue watching the video (some re- buffering)

• Low – the observed quality of the video is not acceptable and the user would willingly stop the video (significant re-buffering)

The test scenario consisted of different experiments described in Figure 5. The same video was used in all the test cases. Note that most OTT video streaming applications do not continuously transmit data; therefore using a GBR bearer would be wasteful of network resources. As a result, non-GBRs are better suited to the usage model.

Figure 5: Description of tests to verify the impact of QoS on Netflix during congestion

Baselining of Netflix performance in a best effort network

A baseline test was conducted to understand the ideal behavior of the Netflix application in a congestion free environment using a default bearer with QCI 9 QoS treatment (see Figure 3).

Subsequent tests introduced congestion and different priority levels.

Test Scenario Test Description

Base line

a. Netflix application as a best effort user / no congestion

b. Netflix application as a best effort user / with congestion

High Priority Netflix application as a high priority user / with congestion

Medium Priority Netflix application as a medium priority user / with congestion

(8)

Netflix session as a best effort user, no network congestion

Figure 6: Netflix throughput, best effort user, no congestion in the network

From Figure 6 it can be observed that

• the Netflix application buffer is filled completely by the initial video stream data chunk which contained 4 MB of data transmitted over a period of 5 seconds

• the initial video stream data chunk contained enough data for the Netflix application to play 40 seconds of video before the application requested for more data to be sent

• the Netflix application buffer was re-filled subsequently by video stream data chunks containing approximately 1 MB of data transmitted every 15 – 20 seconds

• the video quality is high and the stream did not experience re-buffering.

Netflix session as a best effort user with network congestion

A second baseline test was conducted to understand the behavior of the Netflix application in a congested environment using a default bearer with QCI 9 QoS treatment (see Figure 3).

Figure 7: Netflix throughput, best effort user with congestion in the network

Time

Data Re-Transmissions

Throughput

High Quality Medium Quality Low Quality

(9)

From Figure 7 it can be observed that

• the Netflix application buffer is never filled to a point that the application stops requesting data due to lower throughput over the entire session starving the buffer. There were no gaps in transmission as experienced in the previous test

• Netflix user throughput (in red) decreased as network congestion increased

• there was a significant amount and frequency of data re-transmissions (green) due to packet loss

• as congestion reached its maximum level, multiple video stalling / re-buffering events occurred with significant duration

• the Netflix video quality significantly decreased from the initial start as network congestion was increasing.

Netflix session as a high priority user with network congestion

Figure 8: Netflix throughput, high priority user with congestion in the network

Figure 8 shows that

• the Netflix application buffer filled twice to a point where the application stops (similar to gaps in transmission as experienced in the first scenario) requesting data due to the higher available throughput during the initial third of the session as congestion developed

• there was a significant reduction in the amount and frequency of data re-transmissions (green) due to packet loss

• the Netflix application with high priority was able to sustain good video quality at higher congestion levels compared to the best effort user case with congestion. Higher priority increased the overall throughput for the user resulting in the Netflix application maintaining better video quality

•as congestion increased to very high levels, compared to the best effort case, it was observed that video quality did become extremely patchy, but the video stream did not suffer from stalling / re-buffering which had happened in the best effort congestion case.

Time

Throughput

High Quality Medium Quality Low Quality Buffer Full

Buffer Full

(10)

Netflix Session as medium priority user with network congestion

The fourth and final test we conducted to understand the behavior of the Netflix application in a congested environment using a default bearer with QCI 6 (see Figure 3), medium priority QoS treatment. Does the use of QCI 6 and medium priority improve the user experience in congestion situations compared to best effort? Can the application still benefit from some level of priority where minimal extra resources are utilized to assist the session?

Figure 9: Netflix throughput, medium priority user with congestion in the network

Figure 9 contains the measurement. It can be seen that

• the Netflix application buffer filled three times to a point where the application stopped (similar to gaps in transmission as experienced in the very first scenario) requesting data due to the higher available throughput during the initial start of the session and subsequently as congestion was developing. The video codec shifted down to a lower quality, therefore a smaller buffer size was needed

• the throughput of the Netflix application was lower than in the high priority user case but still better than in the best effort case in the congested network

• even during congestion the data download is steady. Retransmission frequency and amount were reduced compared to the best effort case in congestion

• video quality got extremely patchy at higher congestion situations but stalling of video was minimal compared to the best effort congestion case

• medium priority does benefit the user experience and pushes the point where low quality occurs to higher levels of congestion in comparison to the best effort congestion case.

Time

Throughput

High Quality Medium Quality Low Quality

Buffer Full Buffer Full

Buffer Full

(11)

Conclusion

OTT video streaming usage is increasing and is projected to account for vast majority of data transmitted over cellular networks. As an application, video is very demanding for the network while user expectations for quality are equally high.

Netflix can be considered as a proxy for many OTT video streaming applications. Netflix suffers video quality degradation and re-buffering when its data is treated like all other best effort data during periods of congestion. Other video applications which behave like Netflix may benefit as well from longer scheduling length and higher scheduling frequency in order to prevent the application’s buffer from experiencing starvation and lower video quality.

Real world results will vary from controlled lab environments; however the measured effects should be very similar when QoS mechanisms are invoked. Nokia Siemens Networks LTE lab tests point out that the introduction of even medium priority can positively impact the Quality of Experience for the end user during congestion by reducing video quality degradation and re- buffering occurrences. High priority produces the best user experience with better overall video quality in congestion conditions as compared to medium priority but does require the assignment of more network resources to the video streaming session. In congested networks video

streaming applications such as Netflix suffer from reduced video quality and re-buffering even with adaptive codec changes and application buffering.

Figure 10: Test results overview

In a nutshell, our test measurements show that the application of QoS differentiation brings significant improvements in customer experience for video streaming. QoS differentiation is a strategic tool for operators which can be used to develop new business models in content delivery. It is an opportunity for operators to provide a value add that can be monetized.

Case End user experience

1. Netflix streaming as a best effort user / no network congestion

• HD-like video

2. Netflix streaming as a best effort user / DL throughput congestion

• Video quality degraded gradually

• Numerous instances of varying episodes of stalling / re-buffering

3. Netflix streaming as a high priority user / DL throughput congestion

• Video maintained high quality at lower congestion levels

• Medium quality at higher congestions levels compared to no priority case

• Video became extremely patchy but no stalling observed

4. Netflix streaming as a medium priority user / DL throughput congestion

• Video degraded gradually at a similar congestion level as in high priority case

• The throughput experienced by the application was lower and video quality degraded earlier than the high priority case

• Better quality than in no priority case but lower than in high priority case

(12)

Abbreviations

3GPP Third Generation Partnership Project BE Best Effort

eNB Evolved NodeB

GBR Bearer with reserved Bitrate Resources ISP Internet Service Provider

OTT Over-the-Top

LTE Long Term Evolution

Non-GBR Bearer without reserved Bitrate Resources QCI Quality Class Indicator

QoE Quality of Experience QoS Quality of Service UE User Equipment