Measurement-based, Practical Techniques to Improve 802.11ac - - PowerPoint PPT Presentation

Measurement-based, Practical Techniques to Improve 802.11ac Performance

Apurv Bhartia, Bo Chen, Feng Wang, Derrick Pallas, Raluca Musaloiu-E, Ted Tsung-Te Lai, Hao Ma

Nov 2017

ACM Internet Measurement Conference 2017

• Wireless networks have been undergoing changing
• more capacity, higher speed, denser networks
• new 802.11 standards have been introduced
• New techniques are required
• take advantage of the wireless evoluCon

Why is this talk important?

1

What can we do about it?

• Large scale wireless measurement study
• shows changing trend in today’s wireless networks
• TurboCA - A new channel planning algorithm
• reduces wireless TCP latency by up to 40%
• FastACK – A TCP over wireless enhancement
• increases wireless throughput by up to 38%

Agenda

2

• Large scale wireless measurement study
• shows changing trend in today’s wireless networks
• TurboCA - A new channel planning algorithm
• reduces wireless TCP latency by up to 40%
• FastACK – A TCP over wireless enhancement
• increases wireless throughput by up to 38%

Agenda

2

• A wireless standard published in Dec 2013
• wider channel width (up to 160 MHz)
• more antenna usage (4x4 MIMO)
• higher data rate (up to 3.39 Gbps)
• Rapid adopJon of 802.11ac
• default in new wireless devices
• more than 50% of Meraki APs
• 45.7% of the clients devices

What is 802.11ac?

50 100

802.11 abg only 802.11n 802.11ac

% of client devices 2015 2017

27.7% more clients support 802.11ac

3

Measurement Framework

More than one Million Networks Coud Management

4

100K active APs
 
 1.7M associated clients
 
 50G packets

Client Device Trends (2015 – 2017)

2015* 2017

50 100 20MHz only 40MHz 80MHz

% of client devices

Channel Width

50 100 No MIMO 2x2 MIMO 3x3 MIMO 4x4 MIMO

% of client devices

MulJple Antennas (MIMO)

45.7% supporCng 80MHz now Significant usage of 2.4GHz devices

* Large-scale Measurements of Wireless Network Behavior, SIGCOMM 2015

MulC-antenna devices gaining tracCon

50 100 2.4GHz only 2.4GHz + 5GHz

% of client devices

Band

5

Spectrum Trend

2.4GHz is much more congested than 5GHz

0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1

CDF Channel uJlizaJon 5GHz 2.4GHz

1.7% 15%

0.2 0.4 0.6 0.8 1 20 40 60 80 100

CDF Number of interferers 5 GHz 2.4 GHz

5 7 Median # of interferers is similar for both bands

6

Traffic Analysis

0.1 0.2 0.3 0.4

2 8 32 128 256 512 1024

DistribuJon of Packets Bit rate (Mbps)

0 2 8 32 128 256 512 1024

DistribuCon matches types of clients seen in the field

7

More and more 802.11ac devices
 
 Wireless spectrum getting more crowded


• Large scale wireless measurement study
• shows changing trend in today’s wireless networks
• TurboCA - A new channel planning algorithm
• reduces wireless TCP latency by up to 40%
• FastACK – A TCP over wireless enhancement
• increases wireless throughput by up to 38%

Agenda

8

• 802.11ac introduces extended channel width
• channel width is manually configured lower with 35% of the Meraki 802.11ac APs.
• 802.11ac makes RTS/CTS enforced
• RTS/CTS changes the behavior of neighbor APs on the same channel

TurboCA – Motivation

9

• Traffic trend indicates frequent channel switch
• Channel switch disrupts traffic flow

TurboCA – Motivation

One day Snapshot of an AP at Meraki HQ

10 20

480 960 1440

AcJve Clients

4 8

480 960 1440

Usage (Mbps)

50 100

480 960 1440

UJlizaJon (%)

0:00 8:00 16:00 24:00 0:00 8:00 16:00 24:00 0:00 8:00 16:00 24:00 10

• Carrier sensing based channel model
• User-experience based opJmizaJon
• AP load
• channel quality
• channel switch
• Aggressive channel re-evaluaJon to catch up with environment change
• Details in the paper

TurboCA – System

AP load Channel quality channel switch 11 Metric of network Metric of single AP

• Test networks
• museum (171 of 201 APs supporCng 802.11ac)
• university (653 of 735 APs supporCng 802.11ac)
• Comparison
• baseline (default before Nov. 2016)
• TurboCA (default ader Nov. 2016)
• DuraJon
• 03/25/17 -- 05/07/17

TurboCA – Evaluation

Real-world deployment

12

• Usage (throughput)

TurboCA – Evaluation

200 400 600

Per Hour Usage Peak Hour Usage Usage(GB)

University Usage

20 40 60 80 Per Hour Usage Peak Hour Usage Usage(GB)

Museum Usage

Baseline TurboCA Uplink saturated for the university from ISP constraint TurboCA improves peak hour usage by 27% in the museum

13

0.2 0.4 0.6 0.8 1 1 10 100 1000 CDF Wireless TCP Latency (ms) Baseline TurboCA

• TCP latency

TurboCA – Evaluation

18.6ms 31.1ms

40% lower

TurboCA reduces the TCP latency by 40%

14

TurboCA
 
 Better channel assignment

• Large scale wireless measurement study
• shows changing trend in today’s wireless networks
• TurboCA - A new channel planning algorithm
• reduces wireless TCP latency by up to 40%
• FastACK – A TCP over wireless enhancement
• increases wireless throughput by up to 38%

Agenda

15

• TCP is most popular protocol for transmission
• But TCP proposed for wired networks
• not aware of the variable latency introduced in wireless
• not aware of the aggregaCon used in 802.11ac

FastACK– Motivation

AP TCP sender TCP receiver

Key Insight 802.11 latency << TCP latency strong hint for correct recepCon

TCP packet 802.11 ACK TCP ACK TCP ACK TCP ACK 16

`

• Exploit 802.11 ACK hints for larger aggregates
• TCP end-to-end semanJcs maintained
• No client modificaJon
• Details in the paper

FastACK – System

AP TCP sender TCP receiver TCP packet 802.11 ACK TCP ACK TCP ACK TCP ACK TCP ACKs 17

• Testbed – 40, 3x3 clients
• Ixchariot is used to measure the throughput

FastACK – Evaluation

AP2 Switch 2 TCP sender 2 Switch 1 TCP sender 1 AP1 85 feet 30 feet 40 wireless clients

18

• Single-AP mulJple clients test

FastACK – Evaluation

Up to 38% aggregate throughput improvements 150 300 450 600 5 10 20 30 40

Aggregate Throughput (Mbps) Number of Clients Baseline FastACK

19

• MulJ-AP mulJple clients test

FastACK – Evaluation

FastACK does not suffer in isolaCon Combined performance is 60% beker 100 200 300 400 Baseline Baseline FastACK FastACK

Aggregate Throughput (Mbps) Number of Clients

AP1 AP2

251 Mbps

1,2 1,2 1 2

325 Mbps 395 Mbps

20

FastACK
 
 Improved TCP over wireless

s Related Work

• Wireless network measurement

Vivek, S [NSDI 11], Ratul M [SIGCOMM 06], Yu-Chung, C [SIGCOMM 06], Sanjit, B [SIGCOMM 15] The largest scale of study for wireless and first evaluaJon on 802.11ac.

• Channel assignment
• Apurv. B[MOBIHOC 16], Paramvir, B[SIGCOMM 09], Shravan R [MOBICOM 09]

TurboCA considers real enterprise challenges.

• TCP enhancement
• A. Bakre [ICDCS 95], Ajav V [MLICS 95], Hari B [MOBICOM 95],

Stefan S [SIGCOMM 99] FastACK looks at the impact of 802.11ac aggregaJon for TCP.

21

Conclusion

• We observe interesJng trends in today’s wireless networks
• rapid adopCon of 802.11ac-enabled devices
• network densificaCon conCnues to increase
• wireless spectrum geqng more crowded
• Based on observed trends, we present two fundamental techniques
• TurboCA, a new channel assignment algorithm
• FastACK, a TCP enhancing technique over 802.11ac
• New techniques show significant potenJal in enterprise networks

22

Q & A