Emmanuel Weber

10/15/2014

Optimizing Real Time Interactive Video Delivery from the Cloud

Highlights

Cloud based video conferencing solutions are some of the latest novelties in bidirectional real-time video.

Because of the unprecedented scale of those solutions and new type of deployments, there is an opportunity to revisit old problems and provide some new answers to them. This is particularly true in the area of real-time video delivery.

Problems affecting video delivery

Traditional problems: Bandwidth Loss Jitter Out of order packets Duplicate packets Congestion Problems specific to the “last mile”: Congested Wifi Restricted firewall

Brief history of video deployments

Roughly three phases:

Private controlled network

Hosted

Cloud + over the top (OTT) over Internet

Private / Controlled network

The network is fully controlled by enterprise IT

Can implement QoS techniques and dimension the network in such a way that most problems affecting video quality are solved

Doesn’t work well for interactions with parties outside of the controlled network (difficult to setup call and overall quality concerns).

Private / Controlled network

Private Deployment

Hosted

Same software used in private deployment is used in a remote location often accessed over the Internet.

Software is maintained by a third party.

Solution usually not designed to handle Internet traffic or difficult network.

Somewhat worst of both worlds in term of video delivery. Lose benefit of controlled environment and external party does not necessarily see improvements in terms of video quality.

Hosted Deployment

Data Center

Traditional cloud base solution

Dedicated software designed for Internet traffic.

Solution is distributed geographically in multiple data centers to get closer to endpoints.

Traffic can be routed between data centers to optimize network path between endpoints.

Traditional Cloud Base solution

Data Center

Data Center

Cloud based / CDN type deployment

It is possible to improve the video delivery one step further by breaking down the problem once more.

By pushing some of the functionality as close as possible to the endpoints, overall video performance can be greatly improved.

This can be accomplished by a “transparent relay (TR)” that funnels RTP/RTCP streams.

TR ideally should be located as close as possible to the endpoint. This is very similar in concept to traditional CDN solution for video streaming.

Cloud based / CDN type deployment

TR

TR

TR

TR

TR

TR

TR

TR

TR

Data Center

Data Center

Transparent Relay (TR)

MANE type device

As close as possible to the endpoint

Many deployed across the globe. The shorter the round-trip time between the TR and the endpoint, the better.

Provides several benefits: • Error resilience schemes

• Better bandwidth predictor (early congestion detection)

• Better network control (Ex: avoid overloaded peering point)

• Network analytics

Transparent Relay

Packet Queue

RTX FEC

SRTP

Ban

dw

idth

Es

tim

ato

r

Control

Endpoint Cloud RTP/RTCP

RTP/RTCP

Control

• TR is mostly concerned with RTP/RTCP traffic. No control connection between the endpoint and the TR.

• Control connection between the TR and rest of the cloud software. Used to receive configuration such as payload information, encryption keys etc.

• Some lightweight processing (FEC, SRTP) • Network path between TR and rest of the cloud infrastructure is managed by the cloud

operator.

Error resilience schemes

TR can implement several error resilience schemes

Retransmission (RTX) • Particularly effective due to the low delay between the TR and the

endpoint.

Forward error correction (FEC) • Can be generated relatively cheaply in terms of computational power

at the edge of the network to protect the stream.

Stream shaping • TR can shape the video stream when scalable media is used.

Bandwidth predictor

By being closer to the endpoint, there is less network uncertainty.

Bandwidth estimation using indirect methods like delay detection are more accurate and can react quicker to any congestion.

Bandwidth probing is also more reliable.

Better network control

Cloud provider can manage the network infrastructure dynamically with fewer unknown third parties.

In the extreme case, the TR is on the customer premise. This can provide almost the same reliability as private network deployment.

Connections that uses TCP or HTTPS to go through a restricted firewall are terminated earlier greatly improving TCP performance.

Network Analytics

The TR can provide analytics on the quality of the network.

By reacting to those analytics, the cloud operator can adapt its network infrastructure and improve the overall delivery of the video

• Chose alternate provider or network path to serve some areas

• Route calls to a different geography

• …

Conclusion

Using existing technologies, rethinking the video infrastructure in a cloud based context makes it possible to deliver real-time bidirectional video very efficiently.

Cloud based delivery is a relatively new area. There is plenty of room for improvement.

Emmanuel Weber

eweber@bluejeans.com

Thank You