Large Scale WebRTC Panel - WebRTC Conference & Expo III - Santa Clara
While many deployments anticipate large numbers of users, they generally consider the actual WebRTC event to be with a small group of users. Webcasting, gaming, virtual reality, education and a multitude of other applications will require large to very large groups of users to be in actual real-time simultaneously. This session will examine the issues and potential solutions for using WebRTC for these level deployments.
4. Large Scale Challenges: client
#1 WebRTC B2C platform, in production since early 2012
• Frictionless
– Simple: a Permanent link to be reached
– Any browser: WebRTC & Flash >> the same UX, features
based on 1 million minutes of video call /month
11/30/2013
4
5. Large Scale Challenges: client
• Open: No registration, no need to be friend
>> the right signaling protocol: XMPP
>> to build a Guest mode
>> and Gateways to G+, FB, Y!, VK… with SSO
• Add Full-time Presence to the Web
Chrome App
>> Android Native App
5
11/30/2013
6. Large Scale Challenges: client
• Cross-platform WebRTC video calls:
>> all implementations, all devices
– Firefox, Chrome, Chrome OS Devices
– Android Mobile Web & Native app, Google Glass
from a QRcode
to a video call
in a mobile browser
6
11/30/2013
7. Large Scale Challenges: client
• Security
– WebRTC flows 256-bit AES encrypted
– other communications => HTTPS
• Video & Audio Quality, Battery life: Codecs
– OPUS & VP8: software
– H264
• Firefox started H264 integration
• Could help leverage WebRTC
– First VP8 devices
• Google Nexus 5
• Kindle Fire HDX
7
11/30/2013
8. Large Scale Challenges: client
• Multi-party WebRTC Conferencing:
pushing the limits
– 1 permanent link per user: Dedicated signaling
– 4 parties video conferencing
• High CPU usage on the client
• Mesh network: no MCU
– Chat without Chat server: Data channels
– Screen sharing, Chrome only
8
11/30/2013
9. Large Scale Challenges: net
Network communications
• Flexible, interoperability
– REST/JSON
WebRTC standard does NOT
include the signaling
• XMPP over HTTP
• LibJingle
• NAT, Firewalls, 3G/4G
– WebRTC => STUN/TURN
– Flash => RTMP, RTMPT
92%
9
11/30/2013
10. Large Scale Challenges: backend
Server side
• Automated, Scalable, Elastic
Cloud based on AWS EC2
• Gateways: make the contact list run fast
1 gateway activated by 1 user
get list & presence ALL his contacts
+200 Million contacts => Cache Redis
• Geo-localize: to be close to the user
=> DB sharding, TURN clustered
11/30/2013
10