35
Mobile Devices and Wi-Fi Peter Thornycroft March 2014

Mobile Devices and Wi-Fi

Embed Size (px)

Citation preview

Page 1: Mobile Devices and Wi-Fi

Mobile Devices and Wi-Fi

Peter Thornycroft

March 2014

Page 2: Mobile Devices and Wi-Fi

CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved2 #AirheadsConf

Agenda

The commercial value chain

Consumer device reference models

Battery life

QoS

Location

5GHz and DFS channels

Authentication & Passpoint

Handover behavior

Page 3: Mobile Devices and Wi-Fi

3CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Commercial models

• What we see:

– The chain leads to the

cellular operator and

consumer

• What we want to see:

– Some recognition for the

enterprise user

Consumers (your typical Gen-Y) who

don’t care too much about Wi-Fi

performance at work

Chip vendor incorporates driver, is

really responsible for Wi-Fi

functionality, selling to …

Phone / device vendor who has cost

constraints, won’t waste time on

features not of interest to its biggest

customers who are…

Cellular Operators, for whom Wi-Fi

is a minority interest in the first place

and anyway sell to …

Mobile OS

vendor does

some

influencing

Page 4: Mobile Devices and Wi-Fi

4CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

WLANs differ from home APs

Home AP reference model

A single AP, not doing much of interest

WLAN reference model

Many, APs with same SSID and coordinated,

seamless handover (no DHCP, common

authentication etc.)

- No point in looking for other APs because

there (usually) aren’t any

- Established (~correct) behavior is to hang

onto the AP until the signal is very weak,

then switch to cellular data if available

- There is always a ‘better’ AP

- But the device needs to scan (or use neighbor

report) to be aware of the ‘better’ AP.

Benefits of good WLAN client behavior…

- Devices get higher rates

- Other devices get more airtime, better network

capacity

- Less time on the air - better battery life

- Less mutual (co-channel) interference

Same effects are seen in public places, hot zones – ‘always best connected’ activity in Hotspot 2.0 ph3 groups.

Page 5: Mobile Devices and Wi-Fi

5CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Network reference models

• What we see:

– One dual-band home AP

– “give me battery life, and

keep me connected”

• What we want to see:

– Option for multiple-AP WLAN

The current model is the single-AP home network. In this

framework, the best thing is to hold onto your AP until the

signal is too weak to work, then hope you can switch to cellular

data. Probe requests are a waste of battery life because there’s

only one AP.

We want to see either a dual-model or a more flexible

architecture. Maybe sense that there are other APs in the same

system (spot the neighbor report?) and flip to a multi-AP

algorithm.

Under a multiple-AP network, there is always a really-good

signal (except at the edge). It’s just a question of probing

more often to find the better APs.

But it’s difficult to move device, OS and chip vendors away

from their well-established model. They are wary of breaking

what has taken several years to ‘perfect’.

We’ll also see that consumer APs still don’t offer the advanced

features we incorporated some years ago.

Page 6: Mobile Devices and Wi-Fi

6CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Power Save Modes

sleeping

time

beaconDTIM

Traffic for

you

give sleeping

WMM-PS

beaconDTIM

pkt

Traditional Power-Save

U-APSD (WMM-PS)

pkt

pkt

pkt

pkt

pkt

pkt

pkt

pkt pkt

pkt

buffered

time

DTIM

Page 7: Mobile Devices and Wi-Fi

7CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Battery life

• What we see:

– Minimum possible probing

• What we want to see:

– More probe requests in

WLAN

– Using 11k reports

– U-APSD within a beacon

interval

Mobile devices are usually unaware of better AP signals

because they don’t probe enough.

They don’t probe enough because of an over-zealous focus on

battery life, and a model that has only one AP.

Sometimes when a device has an ‘acceptable’ signal it stops

probing altogether. Later, when it starts to move, it may not re-

enable probing until too late to maintain the connection.

In fact, Wi-Fi accounts for less battery consumption than the

cellular subsystem, and far less than the display or CPU

processing app tasks and GPU.

So our focus is on showing device vendors they can ‘go

passive’… only using the 802.11 radio in receive mode.

‘WFA Voice-Enterprise light’, or a collection of features that

enable the device to be multi-AP-aware without reducing

battery life.

Page 8: Mobile Devices and Wi-Fi

8CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

The mystery of missing smartphone QoS

Android

App Code

(QoS – unaware coder)

Driver & microcode

Multi-level QoS

priority API

(that’s OK)

Parrots the driver

API (that’s not OK)

Can’t spell QoS

anyway so it’s

inconsequential

Wi-Fi air interface

• QoS priority (~WMM)is there if

app developers want to use it

• But… it’s not documented And

anyway… app developers are

not QoS-aware– Socket.setTrafficClass(int value) IPTos

• The OS has a hard time figuring out

the QoS Pri required by each app…

• Thus WMM priority is seldom used in

mobile device apps

Same observations apply to WMM-PS (U-APSD) for intra-beacon power save.

Page 9: Mobile Devices and Wi-Fi

9CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

QoS

• What we see:

– WMM functionality exists in

mobile device OS

– But APIs are arcane

– No documentation or

promotion

• What we want to see:

– Better API support

– Developer guidelines

WMM QoS is enabled through the OS to the chip/driver.

But to invoke a high-priority connection, the app developer

must add some parameters to the commands that open sockets .

App developers are unaware of the need to apply Wi-Fi QoS,

and/or are not informed of the required APIs, and/or are not

technically capable of understanding that aspect of app

programming.

This includes developers of voice and video apps including

those in vertically-integrated companies.

Page 10: Mobile Devices and Wi-Fi

10CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Location (distance) enhancements

RTT “Round-Trip-Time”

A standard (actually two standards

and several proprietary variants)

“802.11k”

Location Track Notification,

Modified (to finer timestamps) in

“802.11mc”

Fine Timing Measurements

Distance Calculations

Measure

with me!

Now

here are

my times

t1, t4

OK, here

t1

t3

t4

t2

Challenges:

- Need to combine/average several

frames to get a good reading.

- Averaging many frames affects

battery life, network capacity

Challenges:

- Measuring to nanoseconds

(speed of light: 1 ft per nsec)

- Setting up circuitry to

timestamp the right frame

- Calibration for time frame

leaves (arrives) at the antenna

Once all four timestamps are in one

place, subtraction and /2 gives time-

of-flight and multiply-by-speed-of-

light gives distance

Got

it

Implementation

In mobile device Wi-Fi chips late

2014

In access points 2015 (early

implementation 2014)

No Wi-Fi Alliance certification >>

may cause interoperability teething

troubles

Accuracy should be 1 – 5 metres,

depending on the number of frames

averaged & underlying hardware

Most useful in line-of-sight, but

better accuracy at longer distances

than RSSI

Many variations possible with

WLAN topologies

d = ((t4 – t1) – (t3 – t2)) * c / 2

Page 11: Mobile Devices and Wi-Fi

11CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Location

• What we see:

– RSSI reports

• What we want to see:

– RTT support

– Raw data for RTT, RSSI

Location and location-based-services have attracted the

attention of many commercial and technical principals across

the industry.

Current development is focused on time-based distance

(mostly Round-Trip-Time) measurements:

- 802.11mc Fine Timing Measurement

- Wi-Fi Alliance Wireless Network Management ++

- In-Location Alliance

Look for RTT announcements and features over the next 12

months.

There is a significant danger that this location technology

reverts to proprietary, closed islands rather than developing

along open, standard APIs.

For example:

- Will raw data be available via OS API calls, or mysteriously

processed within the chip/driver or OS itself?

- Will devices built on different chip families interoperate for

RTT location?

Page 12: Mobile Devices and Wi-Fi

12CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

DFS channels – useful at last!

How many radar triggers?

frequency

insallations

0 / year 5 / hour

Usually none, but in some places

> comfortable

Devices supporting DFS

Apple > 2 years

Intel > 2 years

Samsung > 1 year

Others getting there

Most

WLANs

A few

Special concerns

No active client scanning in DFS

bands because they don’t passive-

scan for radar

- slow AP acquisition

- fixed (eventually) by neighbor

report

5GHz Channel count

13 20MHz channels, no DFS

22 20MHz channels including DFS

Channel strategy

Dot them around?

Use the spectrum!

Page 13: Mobile Devices and Wi-Fi

13CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

5GHz band

• What we see:

– Beginning to favor 5GHz

over 2.4

– Spreading DFS support

• What we want to see:

– Overweight 5GHz bias

– 100% DFS support

About 18 months ago Apple supposedly reversed from

unconditionally preferring 2.4GHz to favoring 5GHz.

Unfortunately the battery-saving imperative (see earlier) means

that when a device has an acceptable signal from its AP, it will

stop scanning for a better one. Especially scanning in other

bands.

This can cause difficulties when the WLAN seeks to move a

device to a different band: it may refuse to scan the alternate

band.

DFS support is improving, now available on all Apple devices

(since iPhone 4S) and many Android (since early 2013: e.g.

Samsung Note, Galaxy S4).

We believe this is a good time to start deploying DFS channels.

Page 14: Mobile Devices and Wi-Fi

14CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Passpoint

Identify a hotspot with

Internet reachability and

friendly authentication

Pre-association discovery

What

have you

got?

T-Mobile

BT

Comcast

Orange…

- Pre-association

- New GAS/ANQP protocol

- Lists service providers

- Acceptable authentication

Authenticate to home SP

T-Mobile BTOrange

Accuris

Aicent

BSG…

Hub

(settlement)

RADIUS

e.g. DIAMETER

WPA2 Options

- EAP-TLS

- EAP-TTLS

- EAP-SIM

- EAP-AKA(‘)

Make a list of available

options, decide which to use

Prioritise account options

T-Mobile home (have SIM)

BT visiting (have pwd)

Orange visiting (have pwd)

Comcast visiting (have cert)

Home AP (not Passpoint)

Local (not Passpoint) hotspot

SPs, phone designers all want a

say

- Distinction between ‘home’

and ‘visiting’ hotspot

- May have different tariffs

- Policy for time-of-day,

location…

ANDSF is a cellular protocol that can pass policy to the device to help it make offload decisions.

Passpoint phase 2 introduces se mi-automatic online sign-up and policy services.

T-Mobile SIM

Page 15: Mobile Devices and Wi-Fi

15CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Authentication

• What we see:

– Beginning to support HS2.0

(Passpoint)

• What we want to see:

– Passpoint with EAP-SIM

everywhere

– SPs supporting Passpoint

Passpoint (Hotspot 2.0, from 802.11u) was released as a WFA

certification in June 2012.

For the following 12 months, while SP and enterprise WLAN

equipment supported Passpoint, you could not purchase a

commercial device that was compliant.

That has changed in the last 6 months (iOS7, Samsung Galaxy

S4). Now, we realize that no SP has deployed a network with

standard HS2.0 support.

Why not?

- Actually, NTT has…

- AT&T stayed proprietary

- Cellular operators (see commercial chain above) have no

incentive to allow others (MSOs) to steal their customers

- The enterprise WLAN vendors are waiting for wider

availability

But it’s time!

Public facing vendors should take AOS 6.4, contact a hub

vendor, fire it up and advertise support.

Page 16: Mobile Devices and Wi-Fi

16CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Current handover narrative

A

Good signal, this is dandy!

Time / distance

0 sec

Signal Strength

Page 17: Mobile Devices and Wi-Fi

17CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Current handover narrative

A

Good signal, this is dandy!

OMG, the signal is getting

really low!

Time / distance

0 sec ~30 sec

Signal Strength

Page 18: Mobile Devices and Wi-Fi

18CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Current handover narrative

A

Good signal, this is dandy!

OMG, the signal is getting

really low!

SOS, sending 10 probe

requests on 3 channels

Time / distance

0 sec ~30 sec 35 sec 38 sec

Signal Strength

Page 19: Mobile Devices and Wi-Fi

19CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Current handover narrative

A

B

C D

E

Good signal, this is dandy!

OMG, the signal is getting

really low!

SOS, sending 10 probe

requests on 3 channels

Wowza, responses from 20

APs, how to choose?

Time / distance

0 sec ~30 sec 35 sec 38 sec

Signal Strength

Page 20: Mobile Devices and Wi-Fi

20CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Current handover narrative

A

B

C D

E

Good signal, this is dandy!

OMG, the signal is getting

really low!

SOS, sending 10 probe

requests on 3 channels

Wowza, responses from 20

APs, how to choose?

Let’s reauthenticate with

this one!

Time / distance

0 sec ~30 sec 35 sec 38 sec 40 sec reauthentication request

40.2 sec reauthenticated

Signal Strength

Page 21: Mobile Devices and Wi-Fi

21CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

802.11 k, v, r

• Many features, most important are:

• Neighbor report from AP to client (802.11k)

• Channel report from AP to client (802.11k)

• Beacon report from client to AP (802.11k)

• BSS Transition Management from AP to client (802.11v)

• Fast Transition by client (802.11r)

• (All rolled up in 802.11-2012, 2014)

Page 22: Mobile Devices and Wi-Fi

22CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

802.11r fast BSS transition

C D

R0 key

C

802.1X

authenticator

R0 key

S0 key

S1 key

PTK

Initial Authentication

establishes level 0 keyWLAN distributes

level 1 keys

R1 key

C D

On reassociation,

client presents level

1 key to new AP

R1 key

PTK

S0 key

S1 key

PTK

Mobility domain: A group of

APs covered by a level 0

keyholder

Over-the-air reassociation

widely adopted, over-the-DS

reassociation (via the current

AP) not used

Key suite includes:

Level 0 key (derived at initial authentication,

never exposed OTA)

Level 1 key (per-AP keys) used to derive…

Pairwise temporal keys (to encrypt

communication)

PTK

R1 key

Differences between FT and OKC? … Not much

keyscope keyscope

Page 23: Mobile Devices and Wi-Fi

23CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

802.11k, v, r features

B

C D

E

Neighbor report

AP chan secy key beacon

scope offset

B 6 WPA2 0 45

D 52 WPA2 0 12

E 161 WPA2 0 74

Information about other APs to help

with handover candidate discovery

C

Beacon report

Client reports how it hears (RSSI)

the beacons of other APs

BSSID RSSI

AP B -65

AP D -72

AP E -65

C

BSS Transition Management

AP instructs client to move to

another AP

Move to AP D…

E

DB

D

C

Channel report

AP informs client of channels used

by the WLAN

Channel

6

52

161

Overlaps with neighbor report

Page 24: Mobile Devices and Wi-Fi

24CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

802.11k Neighbor report

• Advertised by AP in the beacon (for all clients, non-

associated) and sent solicited per-client

• List of ‘neighbor’ APs with same SSID includes:

– BSSID

– Channel

– Beacon time offset

– PHY type

– QoS capability

– ‘Key scope’ for common authenticator

• 802.11 does not require neighbor list to be cropped or

ordered or modified per-client (but infrastructure may do so)

• Eliminates the need for active probe request-response

scanning

Page 25: Mobile Devices and Wi-Fi

25CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

The evils of active scanning

• Takes time

– Need to probe on each selected channel in turn, wait ‘reasonable’ interval for responses

– Need to return to current channel for beacon (DTIM)

• Inaccurate results

– RSSI of a single probe response varies ~ +/- 6dB from ‘average’

– Some APs will miss probe requests, or responses are lost

– If the device returns to current channel after ~15msec, sometimes misses responses

• Consumes power

– Typical pattern is to send 2 probe requests per channel, stay awake ~15–20msec

– Each probe request generates ~6 probe responses in a ‘typical’ WLAN

– Each probe response needs an ack

• Consumes airtime, affecting others’ performance

– Frames are sent at low rates, probe responses are retried

Page 26: Mobile Devices and Wi-Fi

26CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Better handover performance with ‘11k’

Current handover sequence:

1. Figure out it’s time to scan

2. Figure out channels to scan

3. Send probe requests, get responses

4. Identify best AP

5. Reauthenticate to new AP

802.11k handover sequence:

1. Periodically request neighbor report

2. Passive scan for neighbor beacons

3. Note if a neighbor AP is ‘better’

4. Reauthenticate to new AP

Probe requests & responses

Signal strength

Time, distance

Signal strength

Time, distance

Behavior c 1999 Behavior c 2013

Signal strength

Time, distanceNeighbor reports & passive scanning

Behavior c 2014 ?

Page 27: Mobile Devices and Wi-Fi

27CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Signal Strength

Proper ‘11k’ handover narrative

A

Good signal, this is dandy!

Time / distance

0 sec

Page 28: Mobile Devices and Wi-Fi

28CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

B

C

D

Signal Strength

Proper ‘11k’ handover narrative

A

B

C D

E

Good signal, this is dandy!

Check neighbor report

every ~10sec

Identify ‘best’ AP and check

for beacon (passive scan)

Time / distance

0 sec ~10 sec 20 sec 30 secB

C

C

D

Page 29: Mobile Devices and Wi-Fi

29CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Signal Strength

Proper ‘11k’ handover narrative

A

B

C D

E

Good signal, this is dandy!

Check neighbor report

every ~10sec

Identify ‘best’ AP and check

for beacon (passive scan)

Signal is low, but I have

already identified the best AP

Time / distance

0 sec ~10 sec 20 sec 30 secB

CB

C

D

C

D

Page 30: Mobile Devices and Wi-Fi

30CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

B

CB

C

D

C

D

D

C

Signal Strength

Proper ‘11k’ handover narrative

A

B

C D

E

Good signal, this is dandy!

Check neighbor report

every ~10sec

Identify ‘best’ AP and check

for beacon (passive scan)

Signal is low, but I have

already identified the best AP

Reauthenticate

Time / distance

0 sec ~10 sec 20 sec 30 sec 30 sec reauthentication request

30.2 sec reauthenticated

Page 31: Mobile Devices and Wi-Fi

31CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Client Match

Client Match forms a virtual Beacon Report:

1. APs measure RSSI from client

2. APs receive beacon reports from the

client

3. Estimate the ‘best’ AP

4. If client is _far_ from ‘best’ AP…

5. Redirect (force handover) to ‘best’

AP

B

C D

E

A

track

-50

-60

-70

-80

AB E

Signal strength

distance

Page 32: Mobile Devices and Wi-Fi

32CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

If 11k, why Client Match ?

• ‘11k’ makes information available to the client

– Neighboring APs, channels, beacon offsets…

• ‘11k’ cannot confirm that the client receives information or how it prioritises

the information

– Neighbor report information may not be used

• Transmitting (or receiving) ‘11k’ does not guarantee that the client will act on

the information

– Handover decisions may not be improved

• Client Match uses information from the infrastructure and the client (if

supports beacon reports)

– The infra knows more about the client’s situation than the client does

• Client Match completes the task by forcing a handover

Page 33: Mobile Devices and Wi-Fi

33CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Handover

• What we see:

– Not much

• What we want to see:

– More probe requests when

in WLAN

– Or… use passive 11k

reports

– Reauthenticate with

802.11r or OKC

Most people think inter-AP handovers take ~1second.

In fact, inter-AP handovers take 30msec, or 250msec, or 7sec

depending on the syndrome.

7sec outages occur when a device (not probing) does not

realize until too late that the signal from its serving AP is

dropping fast. By the time it starts to probe, it has lost the AP

and has to go into cold-start mode. More frequent probes (or

using passive measures as above) would eliminate 7 sec

outages.

Full WPA2 MSCHAPv2 re-authentication takes 200-250msec

to exchange ~50 frames (including acks). This is a stable

figure in the absence of very weak signals due to poor choice

of target AP (mobile devices usually make good AP choices

when aware of their environment through probing). This

outage will be barely noticeable to the user.

But faster re-authentication is possible, through old-school

OKC (from 802.11i) or 802.11r (now available on iPad).

… The ‘bad’ handover syndrome can be solved if the mobile

device is more aware of its surroundings (neighbor report) or

responds to BSS transition management frames (directed

handover from the AP).

Page 34: Mobile Devices and Wi-Fi

34CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved#AirheadsConf

Aruba Utilities shows behaviour

• What we see:

– Frequent long outages

around handover events

• What we want to see:

– More awareness of

environment

– Faster reaction to losing

signal

Aruba Utilities shows very graphically what goes on

when a mobile device moves around an enterprise

WLAN.

Page 35: Mobile Devices and Wi-Fi

35CONFIDENTIAL

© Copyright 2014. Aruba Networks, Inc.

All rights reserved

Thank You

#AirheadsConf