ch11

Question 1

Performance struggles with video streaming

Understand the problems

Answer 1

• Unsupported formats (Installing specific software for each format)
• Missing plugins (FLASH, etc)
• Long start-up times (Only once things are loaded, the video is played. The threshold for people to stay on page is low)
• Frequent stalls (As the video is playing, the video sometimes stops)
• Low quality (HD not available/possible)

Question 2

Dynamic Adaptive Streaming over HTTP (DASH)

What it is and what the concept is

Answer 2

Take the best video and create multiple encodings with different resolutions and bitrates
Split the video into segments of a few seconds
Put all the segments on a HTTP server
Write a manifest that contains the location of each segment
As the bandwidth increases, better quality segments can be retrieved from the server

Question 3

Push vs. pull

Understand the differences (pros and cons)

Answer 3

Push-based video streaming

• “Old” streaming protocols (UDP packets)
• Server pushes packets/stream to client, client acknowledges
• Adaptation based on number of packets received
• Continuous flow from server to client

+ Good performance

• Firewalls and routers block/drop UDP packets
• Therefore often required very specific configurations
• Stream always comes from the same server
• And the server is specifically dedicated for video streaming (high effort) (not scalable)

Pull-based video streaming

• Used in DASH
• Client asks for video segments from server
• Each segment may come from different server (flexible)
• Client reassembles the segments
• Server is doing less work, state-less (scalable)

+ Use of existing infrastructure
+ HTTP traverses firewalls and router much more easily
+ On-demand, live and time-shift streaming. Flexibly used for any type of streaming
+ Can dynamically insert ads
- Performance is less good
- Video players compete with one another for the bandwidth (shifting quality when more players are running)

Question 4

Performance modeling of DASH players

Understand the idea

Answer 4

When multiple players are running at the same time, they will compete unfairly for the network, because they don’t communicate/ see one another.

Can use Dane to coordinate this better by asking for specific bitrates

Better resolution / bitrate (Higher is better)
fewer quality switches (lower is better)
Better Fairness (closer is better)

Modelling:
 - Video players start randomly
 - k different types of users
     \+ Screen size
     \+ Priority
     \+ Video duration

We use a Markov model
Limit the number of players dependent on the minimum necessary capacity of all users
expected number of players = Probability of being in a specific state * number of players in that state
expected video bitrate per device = (Expected no. of players in a state * video quality in that state) / expected no.players in state
Expected number of quality switches = prob of state * probability of switching to state with different bitrate * number of players
that will make the switch