A talk on building applications with magical user experiences on cellular mobile networks. We discuss an application protocol developed within Uber to make mobile experiences realtime and responsive. Find me @ hackinghabits.com

2. Magical User Experiences On
Cellular Mobile Networks
Marketplace Platform
❏ Madan Thangavelu
❏ Nilesh Mahajan

3. 01 Introduction to Uber’s Marketplace Platform
02 Uber app experience challenges
03 How did we solve these problems?
04 Insights
Agenda

4. Introduction to
Marketplace
Platform
Team

5. Goal
Extensible platform for reliably
moving people and things

6. Real-Time
Communications
Geographical
Awareness
High Global
Availability
Logistics
Modeling
Open Platform
Development
Problem areas we work on

7. “Building
systems at
Uber-scale”

8. Rider/Core Flow
Home Enter Destination Request Ride Dispatch Driver Enroute
Shopping Ordering Processing Fulfilling

9. State Machines
Rider State Machine Driver State Machine

10. Synchronize
Rider State Machine Driver State Machine

11. Uber App Experience Challenges
● Time-to-launch app critical on 3G or slower networks
● Synchronize across multiple apps (driver, rider)
● Over tens of parallel RPC request to load various
features at app launch, all competing for resources
● Network used is usually cellular and not WiFi
● Polling is expensive
● Polling payloads change less frequently
● Best polling frequency of N second.
● Higher data consumption

12. How did we solve
these problems?

13. Key Choices
● Moving from sync to async architecture
○ Use of reactive pattern inside apps
○ Reducing blocking network call dependencies
● Reduce client-server interactions
○ Moving from poll to push

14. Components of a Push System
Trigger Service API Layer Push Server App
Pushes
consumer
When to Push? What to Push? How to Push?Triggers

15. Push Protocols
Long Polling
● Work with our HTTP proxies
● Available in browsers
● Simple client implementations
Server Sent Events
● SSE has fewer reconnect roundtrips,
● Lower latency
Client Server
CONTINUE
CONNECT
MSG
45 Sec
MSG
CONNECT
CONNECT
CONNECT
Client Server
CONNECT
MSG
MSG
MSG
MSG
Long Polling Server Sends Events

16. Leveraging Server Sent Events
● Advantages
○ Simple to implement and understand
○ Works on top of HTTP
● Challenges
○ Unidirectional
○ No Visibility
○ Hard to detect broken connections

17. Introducing Ramen
● Real-time Asynchronous MEssaging Network
● HTTP SSE-based push protocol
● Ensures at least once delivery of a push message
● Optimizes amount of data transfer

18. Ramen Protocol
App Ramen Server
/connect?seq=0
Message1,seq=1
Message2,seq=2
/connect?seq=2
Message3,seq=3
Message3,seq=3
Ramen
Session
Connect
Reconnect
HTTP 200 (text/event-stream)
HTTP 200 (text/event-stream)
Ack seq=2

19. Ramen Protocol continued..
● Heartbeats every 5 seconds to detect connection failure
● Message acknowledgement every 30 seconds
● Push message wire priority
○ High
○ Medium
○ Low

20. How speed up app launch?
● App launch process
○ App establishes a Ramen connection with server
○ App make one request to backend - /app/launch
■ Endpoint returns 1 MTU (1500 bytes)-sized response
■ /app/launch also acts as a async trigger to our push systems
○ Based on back-end user state, server generates tens of pushes
○ Pushes are sent on Ramen connection by their priority
○ Based on data in push, experience is rendered on the mobile side

21. Network Pattern
App API
Gateway
App
API
Gateway
Push
Server
Ramen
Pushes
High
Medium
Low
Trigger
/app/launch
connect
Before After

22. 60% reduction
26% reduction
89% reduction
78% reduction
vs Poll :
vs Ramen v1 :
P99
End-to-End Latency

23. How did we make the entire app more
responsive?
● App launch process
○ App establishes a Ramen connection with server
● On any state machine change
○ Kafka events trigger the push systems
○ Based on type of event and backend state, pushes are generated
○ Pushes are sent over the Ramen connection
● All polling calls were changed to push
● Larger pushes were broken down to smaller pushes
● Enforcement and education about payload sizes

24. Concurrent Connections
1M+
Types of apps
20+ Messages Streamed Per
Second
100K+
Messages Delivered for
Driver app
99%
Rider
Driver
Uber Lite
Uber Eats
Restaurants
Uber Fleet
Uber Freight
Web
Jump Bikes
Insights

25. Thank you !

26. Proprietary and confidential © 2019 Uber Technologies, Inc. All rights reserved. No part of this document may be reproduced or utilized in any
form or by any means, electronic or mechanical, including photocopying, recording, or by any information storage or retrieval systems, without
permission in writing from Uber. This document is intended only for the use of the individual or entity to whom it is addressed and contains
information that is privileged, confidential or otherwise exempt from disclosure under applicable law. All recipients of this document are notified
that the information contained herein includes proprietary and confidential information of Uber, and recipient may not make use of, disseminate,
or in any way disclose this document or any of the enclosed information to any person other than employees of addressee to the extent
necessary for consultations with authorized personnel of Uber.