Transcript TGad January 2010 Report

May 2013
doc.: IEEE 802.11-13/0545r1
WLAN QoE, End User Perspective
Opportunities to Improve
Date: 2013-05-15
Authors:
Name
Company
Address
Phone
email
Veli-Pekka Ketonen
7signal Solutions, Inc.
526 S. Main Street,
Akron, Ohio, USA
+1 330 8618150
veli-pekka.ketonen
@7signal.com
Submission
Slide 1
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
Abstract
Presentation for 802.11HEW SG in Hawaii May 12th-17th
Interim meeting
Includes data from various anonymous networks from
7signal Sapphire automated client device QoE
measurements for performance optimization and
management work
Bottlenecks in current networks and suggestions for
improvements are presented as background for further
HEW SG work
Submission
Slide 2
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
The Key Challenge
Submission
Slide 3
Veli-Pekka Ketonen (7signal)
doc.: IEEE 802.11-13/0545r1
In live networks, high max performance does not
translate to sufficient user experience*
•
•
Capacity, range and end user experienced quality do not any more meet needs
Example from University. Nw. vendor has less impact, network config. higher impact
•
* = these are hourly average values from an area of multiple APs/SSID over a week time
Submission
Slide 4
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
Key findings
Submission
Slide 5
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
#1. Too aggressive rate control,
average retrys often exceed 50% (1/2)
•
With 802.11n products, regularly
30-50% of packets require at least
one retry. Often even more.
Too high rate selected
A lot of retries,
multiplied by MIMO-X factor
High utilization
Lower SNR
More radio retries,
up to 7 times/packet
No air time & lost packets
Capacity and TCP
throughput collapse
Submission
Slide 6
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
#1. Too aggressive rate control,
less aggressive has been proven better (2/2)
•
Past findings from Nokia Networks
EDGE link adaptation (Public, Ref [1])
RX-level during this part of the route: -70 – -88 dBm
Original link adaptation
Optimized link adaptation
Throughput
•
Individual network behavior impacts
also all other nearby networks
802.11ac has much more demanding
requirements than 802.11n
RX-level
•
Suggestions:
•
•
Use clearly less aggressive rate control
Make rate control dynamic, adjusting
based on observed radio conditions (like
continuous Bluetooth)
802.11 to specify proper rate control
schemes, instead of leaving this to vendors
Throughput
•
RX-level
-> Reason for improvement:
•Using more MCS-8 instead of MCS-9
Submission
Slide 7
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
#2 Automated channel control algorithms
need clear improvements
•
•
•
Automated features, like
channel selection do not
work properly
Continuous channel
hopping in the whole
network and no stable
state
Impacts also surrounding
networks
• Suggestion:
•
802.11 to specify more in
more detail requirements
Submission
Slide 8
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
#3. Already available radio settings are not
utilized since their impacts are not understood
•
Radio should more accurately and
dynamically operate its settings, like
–
–
–
–
–
–
–
–
–
–
•
Data rates; supported, default, control
Management/control traffic data rates,
Fragmentation process, MTU
QoS
Ack/block ack schemes usage
Long/short pre-amble configuration
RTS/CTS process
Supported 802.11 standards
Minimum limit for probe response
Load balancing, etc
Suggestions
– Performance management practices
– Automated operation
– 802.11 to specify in more detail
Submission
Slide 9
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
#4. Interference due to lacking channel
coordination and Bluetooth devices
• Channel plans are almost random in
public areas
– Resulting packet loss, jitter
• Proper radio operation in the midterm still requires significantly better
channel plans
• Suggestions
– Better, proven automated algorithms
for channel negotiation
– Cloud based control
– Regulation for allowed channels
– Better industry defaults
– Adhoc channel selection
limited/guided
Submission
Slide 10
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
#5. Too dense beacons load air unnecessarily
• In practice solely 100ms used
globally with 1 Mbit/s as
mandatory rate in consumer grade
APs. This congests air
significantly everywhere
• In 100ms, a person walking full
speed moves ~10 inches (~ 25cm).
Is this dense beaconing necessary?
• Suggestions
• Define default beacon intervals longer, ~300ms
• Dynamic/adaptive beaconing, beacon interval automatically dependent on the
observed time between roaming.
• Consider add few % variance to beacon intervals to avoid continuously repeating
collisions (compare to spread spectrum CPU clocking for EMI reduction)
• Consider impact to power save functionality
Submission
Slide 11
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
#6. Mobile networks interfere 2.4 GHz band
WLANs through 3rd harmonic distortion
• When cellular network indoor antennas are near (30ft/10m) to WLAN APs
and/or clients, they may saturate the receiver with off band signals and
receiver generates distortion product that lands in the 2.4 GHz band
DCS-1800
(EUR, US)
PCS-1900 (EUR) UMTS-2100 (EUR)
UMTS-1900 (US) UMTS-1700 (US)
High power
mobile base
station signal
WLAN
Verified to happen in
live network conditions
High power
mobile base
station signal
Distance appr. 300MHz
=> Harmonic distortion
lands at 300MHz
distance from source
WLAN
signal
Ghost
signal
(noise)
=> Signal-to-noise ratio
degrades in WLAN receiver
and data transfer suffers
1.7 GHz 1.8 GHz 1.9 GHz 2.0 GHz 2.1 GHz 2.2 GHz 2.3 GHz 2.4 GHz 2.5 GHz 2.6 GHz
• Suggestion
– Add mandatory RF band-pass filtering to WLAN radios
– Receiver blocking test to FCC approval
Submission
Slide 12
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
#7. Support for legacy devices (802.11b/a)
seriously degrades benefits of new standards
• Protection mode is “contagious” and
highly inefficient
• Benefits of new standards are
limited if legacy devices are
overprotected. Important especially
in consumer grade equipment.
• Suggestions
– Better industry defaults
• 802.11b not supported
• 802.11a not supported
– Improvements to protection mode
– Prevent/limit protection mode
spreading with required minimum
signal levels
Submission
Slide 13
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
#8. Lacking interoperability may take down
whole network performance
• Introduction of new radio
devices increased average retry
rates to about 70%. Max
network capacity came down at
least 50%
• Suggestions
– More exact requirements needed for
client-AP interoperability
– Live network performance
management capabilities need to
improve. All scenarios cannot ever
be tested upfront.
Submission
Slide 14
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
#9. Modest access point antenna solutions
• Omni-antennas with significant vertical coverage are widely used
• RF energy goes where it should not go and antennas try to receive it from
directions where are no clients
• Lacking antenna sophistication
– More gain towards users would benefit uplink quality
– Lack of antenna directivity creates more interference
• Suggestions
– Down-tilt beam patterns
• Fixed, “normal” antennas
• Electrically adjustable, like in mobile networks
– Wider use of beam steering
Submission
Slide 15
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
#10. Performance Management is
completely missing
• With WLAN networks, commonly accepted fact is that:
“It is not necessary to continuously know what kind of service end
users get from the network.
If we manage to make it work once, there is no need to look after
performance for several months/a year. It will take care of itself
automatically. We will troubleshoot when end users complain”
• This approach fundamentally prevents WLAN becoming a reliable media
• Mobile operators/telecom industry are used to manage networks based on
Meaningful Key Performance Indicators, KPIs, that accurately indicate and
predict User Experience (L1-L7) in the network. These are covered also in
standards. This is a good practice that should be brought to WLAN
• Beyond technology providing the required solutions, data and services, even
bigger change is required in attitudes.
Submission
Slide 16
Veli-Pekka Ketonen (7signal)
doc.: IEEE 802.11-13/0545r1
2-10x improvement available with these already
Results: Controller Automation vrs First Manual Optimization Round
•
•
University campus, dense WLAN network
2.4 GHz downlink throughput improvement
–
–
–
•
(+180%)
(+150%)
(+65%)
Area 1
Area 2
Area 3
2.6Mbit/s vrs. 3.5Mbit/s
2.9Mbit/s vrs. 3.8Mbit/s
2.5Mbit/s vrs. 3.5Mbit/s
(+0.9 MOS)
(+0.9 MOS)
(+1.0 MOS)
Area 1
Area 2
Area 3
3.5Mbit/s vrs. 3.8Mbit/s
3.5Mbit/s vrs. 3.9Mbit/s
2.5Mbit/s vrs. 3.5Mbit/s
(+0.3 MOS)
(+0.4 MOS)
(+1.0 MOS)
Area 1
Area 2
Area 3
9% vrs. 1%
9% vrs. <<1%
7% vrs. 1%
(- 89%)
(> -90%)
(-85%)
Hourly minimum measured downlink throughput values increase 10X
–
17
7Mbit/s vrs. 20Mbit/s
10Mbit/s vrs. 25Mbit/s
12Mbit/s vrs. 20Mbit/s
2.4 GHz jitter daily averages before vrs. after
–
–
–
•
Area 1
Area 2
Area 3
2.4 GHz uplink Voice Quality (MOS grade, max 4.0) improvement
–
–
–
•
(+250%)
(+200%)
(+100%)
2.4 GHz downlink Voice Quality (MOS grade, max 4.0) improvement
–
–
–
•
7Mbit/s vrs. 25Mbit/s
5Mbit/s vrs. 15Mbit/s
8Mbit/s vrs. 16Mbit/s
2.4 GHz uplink throughput improvement
–
–
–
•
Area 1
Area 2
Area 3
Improvement
Area1, Area 3
Submission
0.2 Mbit/s vrs. 2.5 Mbit/s
Slide 17
(~1100%)
Veli-Pekka Ketonen (7signal.)
May 2013
doc.: IEEE 802.11-13/0545r1
Areas looking for improvements based on data from current installations
1. Too aggressive rate control
2. Automated channel control algorithms need clear improvements
3. Already available radio settings are not utilized since their impacts are not understood
4. Interference due to lacking channel coordination and Bluetooth devices
5. Too dense beacons load air unnecessarily
6. Mobile networks interfere 2.4 GHz band WLAN’s through 3rd harmonic distortion
7. Support for legacy devices (802.11b/a) seriously degrades benefits of new standards
8. Lacking interoperability may take down whole network performance
9. Modest access point antenna solutions
10. Performance Management is completely missing
Submission
Slide 18
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
References
[1] Nokia Networks/Jussi Nervola: Optimization of EGPRS Link Adaptation, 2007/01/16
–
–
M.Sc. Thesis seminar presentation
http://www.netlab.tkk.fi/opetus/s38310/06-07/Kalvot%2006-07/nervola_160107.ppt
Network statistics from 7signal network optimization and performance assurance reports
Submission
Slide 19
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
ADDITIONAL SUGGESTIONS
Submission
Slide 20
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
Other input for SG work
Topic
Description
Client power control
In dense networks AP’s have often lower power levels than clients. High amount of high power clients with a lot of traffic (data, control or
management) take a lot of air time and cause unnecessary interference.
Prevent hiding SSID’s
Hidings SSID’s increases utilization. All beacons are anyway sent, just without SSID name. Without SSID names in beacons, Clients need to
probe continuously. Remote possibility of hiding SSID names as it provides so little value. Alternatively implement so that beacons are not sent at
all is SSID is hidden.
Improve DFS implementation
DFS operation is commonly reported to be overly sensitive and prevents using DFS channels in many areas. This needs to change with .ac and
.hew to allow efficient use of 5 GHz band.
Improve client roaming behavior
Some clients use excessive off channel scanning. This sometimes results as QoS Null frame flood that takes air time and increase congestion.
Use adaptive beaconing
When there are no relevant clients nearby for a longer time, transmit beacons less often. This reduces overall spectrum load in the surrounding
areas remarkably. Currently APs beaconing (usually 1 Mbit/rate & 100ms interval) consume a lot of spectrum day and night for no good purpose.
Consider ways to work around the power save mode operation.
Use dynamic fragmentation
Client traffic could start with a smaller packets (with fragmentation) at higher data rate (less airtime, less interference). Once rate control has a
good grip of proper rates for that client packet flow, fragmentation could be gradually removed. Rate control is slow and needs some time to
adapt. In addition to lowering retries, utilization and interference, this would allow some time for rate control to work better.
Beacons and probes should use
higher data rates
Beacons (and probes/probe responses) should not be transmitted with the low data rates. By using higher data rate there, the nearby clients (the
only ones relevant to beaconing process) can receive them and the devices further away not. Rate control still should be able to move to lower
data rates when really needed. This would be easy way to reduce utilization further away from AP.
Cloud control:
Radio Link Test and Radio Link
Control interfaces to APs
For automated central coordination of a large group of individual APs, there should be two new functionalities in APs. Radio Link Test
interface: Without authenticating to AP/network, an authorized (password) wireless device should be able to run some basic active measurements
against the AP. These include at least FTP, UDP and ICMP traffic flows. Radio Link Control interface: Without authenticating to AP/network,
an authorized (password) wireless device should be able to control basic Radio settings in AP. These include at least radio channel and power
level. Cloud control: With these interfaces, especially dense consumer AP installations (& SMB) could be centrally controlled and managed to
provide optimum quality for everyone.
Improved radio MAC based device
type classification
Current MAC address based device identification should be expanded. Currently only manufacturer names are recorded. This would allow
developing better automated/by-default-on QoS control algorithms to infrastructure.
Automated “closed loop” radio
network control, adaptive/SON
Optimally, wireless network should reconfigure automatically (SON, Self Organizing Networks). So far in WLAN results are modest. Channel
change algorithms degrade performance, rather than improve it. Networks have to make decisions and learn from impact of changes continuously.
Accurate data of end user experience is needed for this to work properly. Comprehensive data collection and analytics should drive this process.
Submission
Slide 21
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
DATA COLLECTION
BACKGROUND
Submission
Slide 22
Veli-Pekka Ketonen (7signal)
May 2013
doc.: IEEE 802.11-13/0545r1
Background
• 7signal utilizes its products to optimize and operate critical
WLANs (hospitals, universities, enterprises, manufacturing)
• This process includes continuous collection and analysis of
large amount of performance data, consisting of over 600
different metrics. Optimization takes place by reconfiguration
and pre-emptive changes on network based on the data.
• Data includes
–
–
–
–
Automated client device tests, providing L1-L7 data
Passive L1-L2 packet statistics of all 802.11 air traffic
RF environment data
Spectrum analysis data
Submission
Slide 23
Veli-Pekka Ketonen (7signal)
doc.: IEEE 802.11-13/0545r1
7signal Sapphire consists of three elements
Sonar test servers are located in
in close proximity to application
servers
One Eye unit manages 4-7access
points (indoors)
Centrally located Carat manages
Eyes, provides reports and alarms.
Includes Analyzer software
Radio analysis, radio packet
capture and end-to-end application
measurement
• Monitor
• Measure
• Record
• Report
Submission
Slide 24
• Alarm
• Analyze
• Troubleshoot
• Verify
doc.: IEEE 802.11-13/0545r1
7signal Sapphire data covers all layers 1-7
Synthetic tests (L1-L7)
•
•
•
FTP, PING, HTTP, DHCP, SIP, VOIP
Association/authentication/IP address/test
success rates, delays, throughput, latency, jitter,
packet loss, MOS, data rates, failure codes
60 performance indicators, separately for each
AP/SSID/Sonar pair
Spectrum analysis (L1)
•
•
High resolution (280kHz) spectrum analysis for
ISM band
Historical data over months, interference source
analysis with beam steering, compass direction
data on beams
Troubleshooting tests (L1-L7)
RF analysis (L1-L2)
•
•
Access point settings and capabilities, signal
levels, channels, noise levels
40 performance indicators, separately for each
AP, channel, antenna beam
•
•
•
Remote, manual process for troubleshooting
purposes
Full array of tests may be scheduled manually to
each Eye
Eyes may be assigned to perform the additional
tests without interrupting automated monitoring
process
Traffic analysis (L2)
Full packet capture (L1-L2)
•
•
•
•
•
Radio frame header analysis for traffic flow
between clients and access points
Data rates, retry rates, air congestion, roaming,
frame size, device vendor, statistics for all 802.11
frame types, reason codes and status codes
500 performance indicators, separately for each
client, SSID, AP, band, antenna beam
Submission
•
Slide 25
Remote, manual process for troubleshooting
Full blown remote packet capture and easy export
to packet level analyzer like Wireshark, in case
individual radio packet header content information
is needed
Eyes may be assigned to perform the test without
interrupting automated monitoring process
doc.: IEEE 802.11-13/0545r1
7signal Sapphire Eye, the data collection device
•
•
•
•
•
•
•
•
•
•
•
•
•
A “turbo charged” client device
At times active like end users, at times fully passive
Beam steering technology with low noise amplifiers
Integrated compass
802.11 a/b/g/n support
High resolution ISM band spectrum analyzer
High RF performance design; maximized coverage,
minimized quantity, typically 4-7 AP’s per unit
Standard PoE
Neutral design and white color
Attaches easily to ceiling grid, alternatively wall or pole
Indoor and outdoor versions
Data analyzed at device, only key results to database
Minimal load to network, small test packets at
determined intervals
Submission
Slide 26
doc.: IEEE 802.11-13/0545r1
7signal view on QoS, similar to end users
End user
terminals
WLAN
radio
Network
LAN
wired
network
Site
broadband
connection
Core
switching
network
300x
900x
Radio network and
spectrum analysis
200x
End user device
quality of service in
radio network
200x
2x
7signal
Sonar
software
100x
Access Point
100x
Application
servers
Core Router
Servers
10x
LAN switch
100x
100x
7signal
Eyes
Submission
5x
8x
WLAN controller
Core switch
Active end-to-end
quality of service
assessment from end
user perspective
Slide 27
Server racks
doc.: IEEE 802.11-13/0545r1
Network Optimization Flow
1. Ensure that
WLAN/LAN/WAN is
capable providing high
quality service
• Utilize especially “Active/Synthetic” Eye
measurements (“automated end user”)
• Optimize until target levels have been achieved
2. Make sure that all
clients can utilize the
network properly
• “Passive”, client and AP level measurements
• Optimize network and configure clients to achieve this
3. Maintain
performance at the
target level over the
time
• Follow Key Performance Indicators (KPIs) and SLA
tables and take actions
• Proactive corrections are needed when metrics degrade
Submission
Slide 28