We shipped Firefox OS!

to 23 markets

in 15 countries

more coming

more challenging markets

It often looks like this.

Poor Network Conditions

Slow networks (3G = 2.1G)

Unreliable infrastructure

Overloaded base stations

Weather

Users too!

User research

*Very* aware of cost

Shut off data

Even wifi can cost

big opportunity to differentiate

Full-stack impact

Dropped packets

Request timeouts

Non-functional apps

Waste of user's money

Battery quickdeath

Load on base station

Very unhappy users

Very unhappy operators

Turns out...

Not new problems

Lots of research in the field

Little being applied on the Web

Huge wins possible

Graduate Research

University of Michigan

Graduate research lab on mobile performance

Mostly focused on Android on fast networks

But a lot of the findings apply...

Tail Optimization

Radio chips: high/low power mode

Radio Resource Controls (RRC) timers

High power mode stays on after activity

wasted power: "tail energy"

Fast Dormancy

Tell chip that we're done

Little/no access via chipset vendors

Can infer via power measurement

But variable per chipset/network

Even so, problematic to have that access

"Yo, I'm done."

High Latency

Local passive measurement of cellular load

Complex, requires low-level radio access

Could measure requests on ongoing basis?

Still requires high-level positive patterns

Weak Link

Bars don't tell the whole story

Strong link != low latency

Weak link != high latency

Neither signal throughput well

What to do?

What to deux.

We need to think creatively about how to address these problems in a web-developer-friendly way that quantifiably impacts users.

Web Sweet Spot

Tail energy, weak signal, high latency

App developers shouldn't have to care

Positive patterns for built-in awareness

Or at least make it super easy

Muse - PSU Capstone Project

https://github.com/mmes2/muse

Muse - Purpose

Experiment based on the research our teams did in Brasil, Columbia, Thailand, India

News reader app that targets new smartphone user, mostly offline, commutes a lot.

Muse - Features

Self-configuring (geolocation)

Auto populating

Offline reading

Content skinny

Network condition aware

JUST WORKS

Helper Libraries

Passive network info collector

Network info database

Network analysis

= Predict opportunity windows

More experiments

Non-critical requests

Proactive: piggybacking

Last-network-access timestamp

"Now is good" events

Low-budget latency analysis

XHR option to defer

Other Research

https://wiki.mozilla.org/FirefoxOS/Research

THANKS.