Вы находитесь на странице: 1из 66

Wi-Fi performance measurements

Feb 23rd, 2016


Veli-Pekka Ketonen
CTO, 7signal Solutions

Roadmap
@VPonwireless

Material
Jan at
11th , 2016
go.7signal.com/WLPC2016
Copyright © 2015 7signal Solutions, Inc.
Data collection methods

Ac#ve test Beacon analysis Packet analysis


metrics metrics metrics

Network
Full packet Spectrum
element
capture analysis metrics
counter metrics

2 Copyright © 2015 7signal Solutions, Inc.


Data collection tools
Type of metric Examples

Ac#ve test iPerf, IxChariot, Ekahau Site Survey, AirMagnet Survey, Speedtest
app, nPerf app, 7signal Sapphire, 7signal Mobile Eye app

Beacon analysis/ Ekahau Site Survey, AirMagnet Survey, Fluke Aircheck, Metageek
”Wi-Fi scanners” InSSIDer, Wi-Fi Explorer, Wi-Fi Signal, Acrylic Wi-Fi, CommView,
7signal Sapphire
Packet analysis Omnipeek, Metageek Eye P.A., 7signal Sapphire

Network element Cisco Prime, Aruba Airwave, Netscout (NeKlow, SNMP)


counter

Full packet capture AP in capture mode, Mac, Wireshark, Omnipeek, 7signal


Sapphire

Spectrum analysis Wi-Spy/Chanalyzer, Ekahau Site Survey, Cisco CleanAir, Oscium


Wi-Pry, 7signal Sapphire
3 Copyright © 2015 7signal Solutions, Inc.
* Examples, not a comprehensive list
Wi-Fi Experience Depends on 

more than Radio

4 Copyright © 2015 7signal Solutions, Inc.


Wi-Fi connection setup

L2 Discovery § Beacons, probing process


L2 Open authenPcaPon § Authentication request/response
L2 AssociaPon § Association request/response
802.1X/Radius or
L2 CapPve portal § EAP variants / captive portals
L3-4 DHCP
§ IP address, multi-phase process
L4-7 DNS
§ DNS
L4-7 ApplicaPon traffic
§ Applications, TCP, UDP, HTTP,…
L4-7 Roaming
§ Roaming (layers 2/3)
5 Copyright © 2015 7signal Solutions, Inc.
Performance metrics, active test
Association and DHCP/IP services
802.1X authentication §  Multiple phases
§  Association request –  Phase 1: DHCP discovery
– 
§  Association response – 
Phase 2: DHCP offer
Phase 3: DHCP request
§  EAP authentication –  Phase 4: DHCP acknowledged
– 
– 
Multiple phases
Depends on the EAP variant §  Success rate and time
§  Success rate and time
§  Status/reason codes
DNS services
§  Success rates and delays
§  Primary, secondary, tertiary

h\p://www.soasta.com/blog/page-bloat-average-web-page-2-mb/
6 Copyright © 2015 7signal Solutions, Inc.
Association and 802.1X authentication
1.  Started WPA supplicant
2.  Started scan Scan Pme

3.  Started open authentication


A\ach Open authenPcaPon Pme
4.  Completed open authentication Pme
5.  Started association
AssociaPon Pme
6.  Completed association
7.  Connected
Phase 1
8.  EAP authentication started
–  EAP method proposed by the server Phase 2

–  EAP selected method Phase 3


–  EAP peer certificate validated Phase 4

9.  EAP authentication completed Phase 5

*PracPcal steps which can be reported by a standard WPA supplicant


7 Copyright © 2015 7signal Solutions, Inc.
Performance metrics, active test
Throughput Delay/Ping
§  “Goodput”/transfer speed §  Bi-directional delay
§  TCP, HTTP “connection” §  ICMP Echo
§  UDP “connectionless” §  32B typical size
§  One or multiple connections §  “ms” and success %
§  TCP: Slow start & congestion
control must be considered
§  High latency – high
throughput network requires Web page load
long test time §  Load time and success rate
§  “Mbit/s” and success % §  Average page >2MB

8 Copyright © 2015 7signal Solutions, Inc.


TCP behavior with different devices

CongesPon
control

Slow start

h\p://repository.lib.ncsu.edu/dr/bitstream/1840.4/8268/1/TR-2012-6.pdf
9 Copyright © 2015 7signal Solutions, Inc.
Impact of slow start, iPhone 6

*Test by 7signal
10 Copyright © 2015 7signal Solutions, Inc.
Performance metrics, active test
Voice quality Video quality
§  G.711 (64 kbit/s), G.729 (8 §  WMV, H.262/MPEG-2, H.263/
kbit/s), ETSI 06.10 (13 kbit/s) MPEG-4, H.264/MPEG-4, H.
§  Max MOS varies/codec 265/MPEG-H
§  MOS score algorithms §  MOS score algorithms
–  PESQ - Perceptual Evaluation of –  PEVQ - Perceptual Evaluation of
Speech Quality Video Quality
–  POLQA - Perceptual Objective
Listening Quality Assessment
–  PSQA - Pseudo-Subjective Quality
Assessment
§  R-value
–  Based on ITU E-model (G.107)

h\p://www.tamos.com/htmlhelp/voip-analysis/mosandr_factor.htm
h\p://what-when-how.com/voip/e-model-based-voice-quality-esPmaPon-voip/
11 Copyright © 2015 7signal Solutions, Inc.
Performance metrics, active test
Packet loss Jitter
§  Packets received/packets §  Maximum deviation from
sent * 100% average value
§  Ping, bi-directional §  Unit “ms”
§  UDP, uni-directional
§  TCP, “no packet loss”
§  Unit “%”

h\p://www.embedded.com/print/4017837
12 Copyright © 2015 7signal Solutions, Inc.
Active testing, ports and protocols
§  Test payload for active test
must pass end-to-end
§  Firewalls need to allow
test traffic
§  Port numbers can vary
from 0 to 65535
§  Firewalls, proxies,
application filtering, web
accelerators

h\p://resources.intenseschool.com/ccna-prep-fundamentals-of-tcpip-transport-applicaPons-and-security/

13 Copyright © 2015 7signal Solutions, Inc.


Performance metrics, active/passive
Radio retransmissions Data rate/MCSs
§  802.11: If packet is not §  What share of packets use
ACK’ed, a complete packet certain rate/MCS
retry follows §  Data packets use rate control
§  Radio is persistent, will try up always. More reliable to
to 7-25 times for the same calculate rate usage from
packet, same or lower MCS them
§  Retry % (alt. 1) = packets with §  Rate control adjust rates
retry bit on/all attempted based on quality of link and
packets this indicates well signal
§  Retry % (alt. 2) = What share conditions experienced by AP
of original packets require and client
retry (one or many, no matter) §  Rate control algorithms are
§  Active & passive test not standardized in 802.11
§  Active & passive test

14 Copyright © 2015 7signal Solutions, Inc.


Performance metrics, passive test
CRC/FCS errors Channel utilization
§  Receiver synchronizes to packet §  QBSS based utilization
if the 802.11 header (1 or 6 Mbit/ –  How the AP experiences the
s) CRC is correct air utilization?
§  After synch, whole packet is –  Information in beacon IE
transferred –  802.11e element
§  Frame Check Sequence, FCS at –  255 states
the end
–  Supported increasingly
§  Error rate = packets with correct
FCS/all attempted packets §  Packet capture based utilization
–  Calculate based on packet
§  Some chipsets do not show
capture
other than FCS OK packets =>
cannot calculate error rate % –  How much airtime captured
§  Dilemma: Include or do not packets require vs. whole
capturing time
include bad FCS packets?
15 Copyright © 2015 7signal Solutions, Inc.
Air Utilization: Packet capture based model
Inter Frame ArbitraPon Pme Mandatory IniPal
AIFS (used as an average arbitraPon Pme) Back Off Pme
37-70us/frame 13.5-67.5 us/frame

PLCP training Data packet AIFS Back off


field and DuraPon = size/rate Working assumpPon here is that some
header, 20us back off is always present.

PLCP header/preamble Pme +


Frame transmission Pme +
Air u#liza#on = arbitraPon Pme + back off Pme

Total capture Pme


16
Copyright © 2015 7signal Solutions, Inc.
Performance metrics, passive test
Beacon availability Channel number
§  Passive and active scan for §  Static or automated allocations
beacon capture §  Balanced use of all channels
§  0% or 100% §  Channel #
§  # of BSSIDs/channel

Signal level Noise level


§  Signal level (or SINR level) §  Reported by the driver
§  Static or automated AP power §  Hard to measure accurately
levels with any Wi-Fi chip
§  Surveys §  dBm
§  dBm (dB)

17 Copyright © 2015 7signal Solutions, Inc.


Performance metrics, passive test
Data volume Clients/STAs
§  Total volume §  Number of clients
§  Volume/frame type §  QBSS based (beacon)
§  Volume/client §  Observed traffic based
§  MB/Mbit §  Devices manufacturers/types
–  MAC range/OUI

Frame counts Frame sizes


§  Busiest senders §  Average frame size
§  Busiest channels §  Impact of different traffic types
§  Number of frames/min §  Impact of fragmentation
§  Bytes

18 Copyright © 2015 7signal Solutions, Inc.


Performance metrics, active/passive test
802.11 Status codes 802.11 Reason codes
§  802.11 Association Status §  802.11 De-authentication
codes Reason codes
§  Collected from packet capture §  Collected from packet capture
§  Provide information about §  Provides information on
clients association issues
reasons for clients leaving the
§  For example AP
–  Code 17: Association denied due to
overload of the AP §  For example
§  Listed –  Code 5: Disassociated because AP is
–  https://supportforums.cisco.com/document/ unable to handle all currently
141136/80211-association-status-80211- associated STAs. The access point is
deauth-reason-codes busy, performing load balancing, for
–  http://community.arubanetworks.com/t5/
Technology-Blog/802-11-Reason-Codes-
example.
and-Status-Codes/ba-p/257893

h\p://www.aboutcher.co.uk/2012/07/linux-wifi-deauthenPcated-reason-codes/
19 Copyright © 2015 7signal Solutions, Inc.
Performance metrics
TCP socket failure codes Spectrum analysis
§  TCP specific codes for failed §  Spectrum analysis
connections –  QCA/Atheros Wi-Fi chip integrated
– 
§  For example
– 
TI CC2500 chip
Cisco CleanAir (“Cognio”) chip
–  Connection refused
–  Bandspeed chip
–  Network unreachable
–  Socket read error §  Line graphs
–  Attach to AP failed §  Water wall graphs
–  DHCP failure
–  IP address set failure
§  Density graphs
–  Route remove failure §  3D graphs
–  Host unreachable
§  Fingerprinting
–  Connection timed out

20 Copyright © 2015 7signal Solutions, Inc.


How to organize the data?

21 Copyright © 2015 7signal Solutions, Inc.


Why graphs are REALLY needed?

ANSCOMBES QUARTET:
The same mean, variance, correla#on & regression line
Lesson: Always Visualize Data
h\p://data.heapanalyPcs.com/anscombes-quartet-and-why-summary-staPsPcs-dont-tell-the-whole-story/
22 Copyright © 2015 7signal Solutions, Inc.
Organizing the data

§ Key Performance Indicators (KPI)


–  Test results on time axis
–  Visualizations, trending lines
§ Time averages
–  Often does not make sense to show all samples
–  Time averages improve clarity of the data and help with making
the right conclusions
–  Averages: 1min, 10min, 1 hour, 1 day, 1 week, specific period
§ Aggregations
–  BSSID, SSID, frequency band, channel, service area
–  LAN, Wi-Fi
–  Test end points

23 Copyright © 2015 7signal Solutions, Inc.


Service Level Agreement, SLA

§ Helps to quickly determine is the service working as


required
§ Macro KPIs and SLA tables
§ KPI types, consider the difference
–  Individual test is success or fail, value 0 or 100%
§  Test success rate
–  Individual tests have infinite amount of values, “analog nature”
§  Downlink throughput
§ SLA calculation methods
–  Average vs. threshold
–  Sample count vs. threshold

24 Copyright © 2015 7signal Solutions, Inc.


SLA calculation, average based

Compliancy with different threshold values

SLA% = 21Mbps/30Mbps
= 70%

30 Mbit/s

AVG=21 Mbit/s

SLA% = 21Mbps/5Mbps
=> 100% SLA% = 21Mbps/2Mbps
=> 100%
5 Mbit/s
2 Mbit/s

25 Copyright © 2015 7signal Solutions, Inc.


SLA calculation, sample count based

Compliancy with different threshold values

SLA% = 10/25 samples


= 40%

30 Mbit/s

SLA% = 19/25 samples


= 76% SLA% = 23/25 samples
= 92%
5 Mbit/s
2 Mbit/s

26 Copyright © 2015 7signal Solutions, Inc.


Charting

§ Basic chart types


–  Line graphs, bar charts, pie charts
§ X-Y charts
–  Correlation mapping
§ Averages
–  Geometric average Geo_avg = 3√(x13 Ÿ x23 Ÿ x33)
–  Arithmetic average Arith_avg = (x1 + x2 + x3)/3
§ Min, max, percentile (5%, 95%)
–  Calculated inside the averaging period sample
§ Cumulative Distribution Functions (CDFs)
–  Continuous distribution graph

27 Copyright © 2015 7signal Solutions, Inc.


Impact of sampling density

Event in
Passive test network

Time
Measurement Measurement

Performance
Ac#ve test of network

Time
Measurement Measurement

§ Manual or automated, your test data rarely covers


everything
28 Copyright © 2015 7signal Solutions, Inc.
How to use the data and 

make conclusions?

29 Copyright © 2015 7signal Solutions, Inc.


Collecting the test data

Site Trend Email SNMP


surveys analysis reports alarms

One off SLA Email


tests failure alarms

30 Copyright © 2015 7signal Solutions, Inc.


Drilling in to issue

SSID

Frequency
band

Service
area

AP/BSSID

Segment

Client
vendor

Client

31 Copyright © 2015 7signal Solutions, Inc.


Spotting the problems from data

Missing ProblemaPc
High packet status/reason Too many
beacons/ clients/AP
loss/ji\er codes
dead AP

Failing to High air


High delay Interference uPlizaPon
authenPcate

ConPnuous
Low Failing to channel High retries/
throughput get IP low MCS
changes

No e2e Poor channel


connecPvity allocaPons

Ac#ve test Passive test


32 Copyright © 2015 7signal Solutions, Inc.
Data from a major University in US

•  Network has been already optimized and works well in most


aspects
•  Careful analysis reveals still room for further checks
•  Challenging, open environment
•  HT20 channels used for maximum capacity
•  Example data on observed network performance in the main
student SSID
•  Data from the 1st floor with 8 APs, where a lot of students spend
time
•  2 weeks period around the time school starts

33 Copyright © 2015 7signal Solutions, Inc.


Dashboard Macro KPIs
ConnecPvity IP services
•  Beacon availability •  IP address Pme
•  Radio a\ach success rate •  IP address success rate
•  Radio a\ach Pme
•  AssociaPon success rate
•  Radio associaPon Pme
•  TCP test success rate

AuthenPcaPon Quality
•  Open radio authenPcaPon success rate •  TCP DL throughput
•  Open radio authenPcaPon Pme •  TCP UL throughput
•  Radius: Total EAP authenPcaPon •  Ping RTT
success rate •  Web page load Pme
•  VoIP MOS DL
•  VoIP MOS UL

34 Copyright © 2015 7signal Solutions, Inc.


Dashboard, 1st floor

1 month period

Period of interest
35 Copyright © 2015 7signal Solutions, Inc.
SLA: Connectivity and Authentication

One hour averages

School off School starts 5 GHz doing well, but


li\le below 100%

2.4 GHz struggles


with load

School off School starts 5 GHz doing well

2.4 GHz struggles


with load

36 Copyright © 2015 7signal Solutions, Inc.


SLA: Connectivity and Quality

One hour averages

School off School starts 5 GHz doing well, but


li\le below 100%

2.4 GHz struggles


with load

School off School starts 5 GHz doing well

2.4 GHz struggles


with load

37 Copyright © 2015 7signal Solutions, Inc.


SLA: Table view for 5 GHz

Daily averages

Retries above
Good performance
10% target
level

38 Copyright © 2015 7signal Solutions, Inc.


SLA: Table view for 2.4 GHz

Daily averages

Voice quality UL
below target

Retries above 10%


target level

Radio a\ach not on Throughput not on


target level target level

39 Copyright © 2015 7signal Solutions, Inc.


Traffic volume

40 Copyright © 2015 7signal Solutions, Inc.


Station count and channel utilization

AreaBand average

70 clients/AP at
5 GHz
5 clients/AP at
2.4 GHz

60% uPlizaPon at
2.4 GHz
25% uPlizaPon at
5 GHz

41 Copyright © 2015 7signal Solutions, Inc.


Station count and channel utilization

2.4 GHz/AP specific

Max 12 clients/AP
(AP140)

Up to 80%
uPlizaPon (AP140)

42 Copyright © 2015 7signal Solutions, Inc.


Station count and channel utilization

5 GHz/AP specific

Max 100
clients/AP
(AP156). Others
have clearly less

Up to 90%
uPlizaPon
(AP 102)

43 Copyright © 2015 7signal Solutions, Inc.


Frame volumes, Uplink/Downlink

5 GHz band carries much


higher traffic load/AP

44 Copyright © 2015 7signal Solutions, Inc.


End-to-end performance KPIs

45 Copyright © 2015 7signal Solutions, Inc.


TCP throughput, Uplink/Downlink (active)

Drop

Drop

46 Copyright © 2015 7signal Solutions, Inc.


Voice quality (MOS) and RTT (active)

Peak load
increases
latency a lot at
2.4

47 Copyright © 2015 7signal Solutions, Inc.


Jitter and Packet Loss (active)

48 Copyright © 2015 7signal Solutions, Inc.


Probe request and response air utilization

Requests and
responses in
balance

49 Copyright © 2015 7signal Solutions, Inc.


Beacon air utilization and frame errors

Looking good: Beacons consume


very limited airPme

50 Copyright © 2015 7signal Solutions, Inc.


Retransmissions, AP/client (passive)

AP retries at both bands


around 50%-60%

Client retries at both


bands around 30%

51 Copyright © 2015 7signal Solutions, Inc.


Spotting specific issues

52 Copyright © 2015 7signal Solutions, Inc.


SLA: Connectivity and Authentication

2.4 GHz APs

AP 156 Failing Degraded AP 127 Failing


connecPvity connecPvity connecPvity

AP 156 Failing AP 127 Failing


authenPcaPon authenPcaPon

53 Copyright © 2015 7signal Solutions, Inc.


SLA: IP services and Quality

2.4 GHz APs

Several APs
lacking IP compliancy

Degraded
quality

54 Copyright © 2015 7signal Solutions, Inc.


IP address success rate and time

2.4 GHz APs

IP is always received

…a bit long delay to get IP.


SLA set at 5s.

55 Copyright © 2015 7signal Solutions, Inc.


Reasons for low IP address score

2.4 GHz APs

Significant delay: DHCP


host not responsive

Delay gerng offer:


DHCP host not responsive

56 Copyright © 2015 7signal Solutions, Inc.


Reasons for low connectivity scores 

2.4 GHz: AP127 and AP156 connectivity issues

APs send/beacons
received pre\y normally

AssociaPon
success rate degraded AP 156 & 127 open
authenPcaPon fails

57 Copyright © 2015 7signal Solutions, Inc.


Reasons for low connectivity scores 

2.4 GHz

Part of the APs need clearly


longer Pme for associaPon

58 Copyright © 2015 7signal Solutions, Inc.


Throughputs, daily minimum

Significant slowdown under load, but 5 GHz


remains sPll at mulP Mbps speed

AP140 lowest

Significant slowdown under load, but 2.4 GHz


remains sPll at 1 Mbps speed (AP140)

59 Copyright © 2015 7signal Solutions, Inc.


2.4 GHz channel usage

Vast majority of the


nearby APs use channel 1

Vast majority of the


nearby APs use channel 1

60 Copyright © 2015 7signal Solutions, Inc.


5 GHz channel usage

Vast majority of the


nearby APs use channel 36

In-balanced channel
usage, unused capacity

61 Copyright © 2015 7signal Solutions, Inc.


Reason codes (passive analysis)

“4-way handshake Pme


outs” with mulPple APs

“AuthenPcaPon failed” with


mulPple APs, especially with
AP 140

62 Copyright © 2015 7signal Solutions, Inc.


Plan for action
§ Check DHCP server serving certain APs, like 140, 102-2
–  Impacts both bands
–  Are there two separate DHCP servers being used by measured APs?
–  Allocate more DHCP capacity to speed up responses
§ Restart AP156 and AP127
–  Observe if open authentication issues disappear
–  If not, consider replacing the APs next
§ Balance usage of channels on both bands
–  Helps to reduce throughput drops during high load
–  Helps to perform successful associations and open authentications
§ Improve especially AP140 (2.4 & 5 GHz) RF performance
–  AP140 regularly experiences 60-80% air utilization at 2.4 band
–  AP140 also experiences lowest throughputs with active tests in both
bands
–  “Authentication failed” codes may be due to RF issues
–  => Consider new channel and power settings for better channel
quality and better load balancing between APs

63 Copyright © 2015 7signal Solutions, Inc.


Demo

64 Copyright © 2015 7signal Solutions, Inc.


Cisco, Apple, Xirrus,…

65 Copyright © 2015 7signal Solutions, Inc.


Thank You!

Email: veli-pekka.ketonen@7signal.com
Presentation: go.7signal.com/WLPC2016

www.7signal.com
@VPonwireless

66 Copyright © 2015 7signal Solutions, Inc.

Вам также может понравиться