1. All rights reserved. ©2020All rights reserved. ©2020
Scalable High Efficiency Video
Coding based HTTP Adaptive
Streaming over QUIC
August 14th, 2020
ACM SIGCOMM EPIQ Workshop
Minh Nguyen, Hadi Amirpour, Christian Timmerer, Hermann Hellwagner

2. All rights reserved. ©2020
● Motivation
● Contributions
● Proposed method
● Evaluation and discussion
○ HTTP/3 over QUIC vs HTTP/2 over TCP
○ Proposed method vs state-of-the-art methods
● Conclusion and Future work
Agenda
All rights reserved. ©2020
2

3. All rights reserved. ©2020
● Protocols
○ HTTP/2 suffers from Head-of-line (HOL) blocking.
○ QUIC running on top of UDP can tackle this issue.
● Video streaming
○ Adaptive bitrate (ABR) algorithms are mainly designed
for either non-scalable or scalable video coding
formats.
○ Lack of an approach that works well for both non-
scalable and scalable video coding formats.
Motivation
QUIC
Stream 1
Stream 2
Stream 3
Non-scalable video Scalable video
Quality
FHD
HD
SD
Quality
FHD
HD
SD
SegmentSegment
Enhancement
layer 1 (EL1)
Base layer (BL)
Enhancement
layer 2 (EL2)
Server Client
3
HTTP/2
Stream 1
Stream 2
Stream 3
HOL blocking
Server Client

4. All rights reserved. ©2020
● A systematic comparison of QUIC and HTTP/2 regarding the multiplexing feature.
● A non-scalable video streaming ABR algorithm in combination with an additional
download technique is proposed to not only improve the video quality but also to provide
a smooth adaptation behavior.
Contributions
QUIC HTTP/2
><
Non-scalable
ABR
Additional download
technique
Video quality
Smooth adaptation
Scalable video streaming
4

5. All rights reserved. ©2020
Proposed method
● State-of-the-art ABR algorithms
○ Non-scalable based method: Aggressive ABR (AGG)
○ Scalable based method: Backfilling
● Proposed method for Scalable Video Streaming
○ Modified AGG + HTTP/2-Based Retransmission technique (H2BR)
○ Modified AGG
■ Choosing the number of layers for each segment based on the
network condition,
■ Downloading sequentially from low to high layers of each segment.
○ H2BR [PV’20]
■ Filling quality gaps in the buffer,
■ Downloading concurrently next layers and the additional layer with
priority and multiplexing features,
■ Terminating layers with termination feature.
Scalable based Backfilling
Modified non-scalable based
AGG
Modified non-scalable based
AGG + H2BR
5

6. All rights reserved. ©2020
Proposed method
How does H2BR work?
○ Detecting gaps in the buffer.
○ If there is a gap, additional layer will be 1-level higher
(i.e., the segment has BL, the additional layer is EL 1).
○ Additional layer will be downloaded if the throughput
can sustain the next layer and additional layer so that:
■ Retransmission buffer > 0, and
■ Estimated buffer > BufferSize/4.
○ Assigning priority weights for additional layer and next
layer so that for these layers enough throughput is
allocated.
○ Sending 2 requests.
○ Terminating the additional layer if:
■ Retransmission buffer < 100 ms, or
■ Current buffer < BufferSize/4.
Modified non-scalable based AGG +
H2BR
6

7. All rights reserved. ©2020
Evaluation and discussion
Server
Client
All rights reserved. ©2020
ABR
algorithm
H2BR
DummyNet Tool
4G NetworkLSQUIC for QUIC
Nghttp2 for HTTP/2
LSQUIC for QUIC
Nghttp2 for HTTP/2
Video
EL 3
EL 2
EL 1
BL
Testbed
* BL: Base layer
* EL: Enhancement layer
7

8. All rights reserved. ©2020
Evaluation and discussion
Modified AGG
(M-AGG)
Backfilling
(BF)
All rights reserved. ©2020
Modified AGG + H2BR
(H2BR)
Compared methods
8

9. All rights reserved. ©2020
Evaluation and discussion
Impact of packet loss rate on the performance of adaptation approaches
All rights reserved. ©2020
Average quality level # downward switches
HTTP/3 over QUIC vs HTTP/2 over TCP
9

10. All rights reserved. ©2020
Evaluation and discussion
# additional layers successfully downloaded by H2BR
All rights reserved. ©2020
HTTP/3 over QUIC vs HTTP/2 over TCP
10

11. All rights reserved. ©2020
Evaluation and discussion
Average quality level # downward switches
Impact of buffer size on the performance of adaptation approaches
All rights reserved. ©2020
Proposed method vs state-of-the-art methods
11

12. All rights reserved. ©2020
Evaluation and discussion
Buffer starvation when buffer size is 5s
All rights reserved. ©2020
Proposed method vs state-of-the-art methods
12

13. All rights reserved. ©2020
Conclusion and Future work
All rights reserved. ©2020
● Conclusion
○ QUIC can well support concurrent streams to provide a better
performance in case of packet loss.
○ Proposed method makes non-trivial improvement in scalable
video streaming.
○ H2BR might be a burden that can lead to buffer starvation
when the buffer size is small.
● Future work
○ Investigating parameter selections for H2BR.
○ Considering different network traces and video contents.
13

14. Thank you
All rights reserved. ©2020
14