Quality of Experience and Quality Feedback
Markus Fiedler
Blekinge Institute of Technology School of Engineering
Dept. of Telecommunication Systems Karlskrona, Sweden
HET-NETs’06 Tutorial T13 Ilkley, UK, Sept. 2006
My Own Background (1)
Moved from the network towards the user ☺
• Working with Grade of Service/Quality of Service issues since 1992
– Admission control, dimensioning
• Got interested in end-user throughput perception in 2000
– “Kilroy”-Indicator 2002 co-developed with Kurt Tutschku, University of Würzburg
• E-Government project 2002—2004 – Implications of IT problems
• Preparation of the NoE EuroNGI 2003
EuroNGI-Related Activities
• Leader of
– Joint Research Activity JRA.6 “Socio- Economic Aspects of Next Generation Internet”
– Work Package WP.JRA.6.1 “Quality of Service from the users’ perspective and feedback mechanisms for quality control”
– Work Package WP.JRA.6.3 “Creation of trust by advanced security concepts”
• EuroNGI-sponsored AutoMon project (2005) – Improved discovery of end-to-end problems – Improved quality feedback facilities
My Own Background (2)
• Projects within Intelligent Transport Systems and Services since 2003
– Timely delivery is crucial (dependability, safety)
– Network Selection Box (GPRS/UMTS/WLAN) – How to match technical parameters and user
perception?
• Surprised that rather little attention has been paid to user-related issues by “our” scientific community
Thesis 1:
Users do have – sometimes unconscious –
expectations regarding ICT performance
Quality Problems?!?
Perception of Response Times
100 ms 1 s 10 s Response
time
Reacts promptly
There is a delay
Flow of thoughts interrupted
Un- interesting Boring
• Most users do not care about “technical”
parameters such as Round Trip Time (RTT),
one-way delay, losses, throughput variations, ...
Some User Reactions (1)
• Study by HP (2000) [1]
• Test customers were exposed to varying
latencies when composing a computer in a web shop and had to rate the service quality
• Some of their comments are found below:
• Understanding that there’s a lot of people
coming together on the process makes us more tolerant
• This is the way the consumer sees the
company...it should look good, it should be fast
Some User Reactions (2)
• If it’s slow I won’t give my credit card number
• As long as you see things coming up it’s not nearly as bad as just sitting there waiting and again you don’t know whether you’re stuck
• I think it’s great...saying we are unusually busy, there may be some delays, you might want to visit later. You’ve told me now. It I decide to go ahead, that’s my choice.
• You get a bit spoiled. I guess once you’re used to the quickness, then you want it all the time
Consequences?
[2] summarises:
• 82% of customer defections are due to
frustration over the product or service and the inability of the provider/operator to deal with this effectively
• ... on average, one frustrated customer will tell 13 other people about their bad expeciences ...
• For every person who calls with a problem, there are 29 others who will never call.
• About 90% of customers will not complain
before defecting – they will simply leave once they become unsatisfied.
Shortcomings in
perceived dependability are likely to cause churn!
Quality of Service (QoS)
• Telecom view
– ITU-T E.800 (1994) defines QoS as “the collective effect of service performance
which determine the degree of satisfaction of a user of the service”, including
• Service support performance
• Service operability performance
• Serveability (Service accessibility/
retainability/integrity performance)
• Service security performance
– QoS measures are only quantifiable at a service access point
Quality of Service (QoS)
• Internet view
– Property of the network and its components
• “Switch A has Quality of Service”
– Some kind of “Better-than-best-effort”
packet forwarding/ routing
• RSVP
• IntServ
• DiffServ
• Performance researcher view
– Results from queuing analysis
Quality of Experience (QoE) [2, 3]
• Rather new concept, even more user-oriented than QoS: “how a user perceives the usability of a service when in use – how satisfied he or she is with a service” [2].
• Includes
– End-to-end network QoS
– Factors such as network coverage, service offers, level of support, etc.
– Subjective factors such as user expectations, requirements, particular experience
• Economic background: Dissapointed user may leave and take others with him/her.
Quality of Experience (QoE)
• Key Performance Indicators (KPI)
– Reliability (service quality of accessibility and retainability)
• Service availability
• Service accessibility
• Service access time
• Continuity of service
– Comfort (service quality of integrity KPIs)
• Quality of session
• Ease of use
• Level of support
• Need to be measured as realistically as possible
Thesis 2:
There is a need for more explicit feedback
to make the user feel more confident
Cf. [4]
Section 2.4
Typical
Feedbacks
Types of Feedback
• Explicit feedback
– Positive/negativ acknowledgements
• E.g. TCP
– Asynchronous notifications
• E.g. SNMP traps
• Implicit feedback
– Can be obtained through observing whether/how a process is happening – Dominating Internet as of today
1. Feedback from the Network
a. Network Application
• Implicit: No or late packet delivery b. Network Network Provider
• Classical Network Management/monitoring c. Network User
• Implicit: “Nothing happens...”
• Rudimentary tools available
• Operating system issues warnings Within the network stack: control packets
2. Feedback from the Application
a. Application Application
• Some applications measure the performance of the packet transfer and adapt themselves (e.g. Skype, videoconferencing)
b. Application User
• Implicit by not working as supposed
• Explicit by notifying the user or adapting itself
c. Application Service Provider
• Active measurements of service performance d. Application Network Provider
• Monitoring of control PDUs
3. Feedback from the User
Implicit: give up / go away = churn Explicit:
a. User network operator
• Blame the closest ISP
• Not uncommon ISP attitudes:
• The problem is somewhere else
• The user is an idiot b. User service provider
• Online quality surveys c. User application
• Change settings
4. Feedback from the Service Provider
• Towards the network operator in case of trouble
• Part of the one-stop service concept [4]:
– Service provider = primary point of contact for the user of a service
– User relieved from having to search for the problem (which is the service provider’s business)
The Auction Approach
Cf. [5]
Chapter 5
Feedback Provided by Bandwidth Auctions
a. Bidding for resources on behalf of the user b. Signaling of success or failure
c. Results communicated towards the user
• Successful transfer at resonable QoS
• Unsuccessful transfer at low cost
d. Results communicated to network (and perhaps even service) provider
• Dimensioning
• SLA
The AutoMon Approach
Cf. [5]
Chapter 6
AutoMon Feedback
• DNA (Distributed Network Agent) = main
element in a self-organising monitoring overlay a. Local tests using locally available tools
b. Remote tests and inter-DNA communication
• Comparison of measurement results
c. Alarms towards {network|service} provider(s) in case of perceived problems
• E.g. using SNMP traps
d. Lookup facilities for providers
• E.g. saving critical observations in a local MIB e. Notification facilities towards users
• Not mandatory, but maybe helpful
The AutoMon Project
• Design and Evaluation of Distributed, Self- Organized QoS Monitoring for Autonomous
Network Operation (http://www.informatik.uni- wuerzburg.de/staff/automon)
• Sponsored by the Network of Excellence EuroNGI (http://www.eurongi.org)
• Partners (and Prime Investigators)
– Infosim GmbH & Co. KG, Würzburg (S.
Köhler, M. Schmid)
– University of Würzburg, Dept. of Distributed Systems (K. Tutschku, A. Binzenhöfer)
– Blekinge Institute of Technology, Dept. of Telecomm. Systems (M. Fiedler, S. Chevul)
The AutoMon Concept
1. DNA = Distributed Network Agent – Self-organising
– Prototype available
• Network operations
• Simulations
2. NUF = Network Utility Function – Quality evaluation:
user impairment = f (network problems) – Focus on throughput (TUF)
3. QJudge = Demonstrator for
– Quality evaluation (traffic-lights approach) – Feedback generation (traps)
– MIB
The Way To Autonomous Networks
IT-System e.g. LAN/MAN
Autonomous Manager
Input Output
Autonomous Manager
Autonomous Manager
Act Observe
Analyze
Disadvantages of a Central Monitor Station
NMS
?
?
?
? ?
?
?
Mailserver
Webserver
Client B Client A
Client D
Backup Server Client C
Link status: up down
?
unknownMailserver
Webserver
Client B Client A
Client D
DNS Server Client C
DNA
DNA
DNA
DNA
DNA
DNA
DNA
DNA reroute
Extended view
temporary DNS proxy
Link status: up down
?
unknownAdvantages of Distributed Monitoring
NMS
DNA Phase 1: Local Tests
Cable?
IP?
DNA Ping!
-NIC-Status
-NetConnectionStatus -PingLocalHost
-IPConfiguration -DNSConfiguration -DHCPLease
-EventViewer
-HostsAndLmHosts -RoutingTable
-PingOwnIP
-PingWellKnownHost
DNA Phase 2: Distributed Tests
Server DNA
Ping!
Test please Test
please
DNA
Ping!
Ping!
Result
DNA
Result
-PingSpecificHost
-PingWellKnownHosts -DNSProxy
-RerouteProxy -PortScan
-Throughput -Pinpoint Module
Server
The DNA Overlay Network
Internet
DNA
DNA DNA
Use of a P22-based overlay network
• DHT = Kademlia
• Peer = DNA
DNA
DNA
DNA
DNA
Challenges:
- keep overlay connected
- locate specific DNA - locate random DNA
Scalability Results Using the DNA Prototype
0 200 400 600 800 1000 1200
100 150 200 250 300 350 400 450 500
Overlay size
Average search duration [ms]
Average online time = 60 min No churn
Network Utility Function
Internet
DNA
DNA
U
InU
NetwU
Out= U
Netw• U
Inevaluate original
quality
evaluate quality of the network
evaluate received
quality
Network Utility Function
• Range of U: 0 (worst) ... 100 % (best) – intuitive for – Users
– Providers – Operators
• Captures performance-damping effect of the network
– UNetw = 1 network “transparent”
• Bad service perception (Uout 0) can have its roots in
– Badly performing network (UNetw 0) – Badly performing application (Uin 0)
U
Out= U
Netw• U
InThroughput Utility Function
• Basis: Throughput
– on small time scales ΔT
– during observation interval ΔW
• m-utility function
U
m:captures impact of changes in traffic volume – Overdue traffic ( late or lost)
• s-utility function
U
s:captures impact of changes in traffic burstiness – Shaping = reduction ( throttle)
– Sharing = increase ( interfering traffic)
• n-utility function
U
n:– Bias by network (e.g. UMTS vs. LAN)
U
Netw= U
m• U
s• Un
Recent Skype-via-UMTS results:
PESQ and NUF/TUF [6]
PESQ = Perceptual Evaluation of Speech Quality NUF = Network Utility Function
TUF = Throughput Utility Function
l
SNMP Interface
• Trap generation
– Upon threshold crossing, e.g.
• Green Yellow
• Yellow Red
• (Enterprise) MIB
– Not yet designed
– Cf. RMON history group
• Statistics (m, s)?
• Array with values?
• Histograms?
• Why not just participate in the overlay? ;)
Simple parameters for monitoring of Skype:
U
Netw= U
m≥ 80 %
U
Netw= U
m≥ 50 %
U
Netw= U
m< 50 %
Thesis 3:
The user needs to be relieved from
decisions based on incomplete feedback
Status
Internet usage still implies a high degree of self-service
• Some kind of Internet paradigm (just provide connectivity, the rest is left to the user)
• The “Anything-over IP-over-anything” principle provides both opportunities and nightmares
• Mastered differently by different applications (better by some, worse by others)
• A lot of “decision making” is left to the user – does (s)he really know about the implications?
• Recent trend towards IMS (Internet Multimedia System): might help, but will the Internet
community accept it?
Status
Issues:
• How do subjective QoE and objective QoS parameters match each other?
– More or less solved for some applications
• How can I be sure that
– “my” task is performed and completed
– “my” problems are detected and worked on in time?
• Which network can be used for a particular task?
– Rough indications available
• “Money back” policies?
– cf. airlines and (some) train companies
Solving these issues increases dependability perception and thus trust
Wish-list
• No additional complexity for the user!
– Application of self-organisation principles
• Preventive feedback:
– Clear guidelines and indications regarding (im-)possibilities
• Optional cross-layer interfaces required
• Reactive feedback:
– Signalling of success or failure
• Again a matter of cross-layer interfaces – Action on behalf of the user
• Notifications
• Selections (e.g. a particular network)
Wish-list (continued)
• The Internet community should care about end user perception – tendencies visible:
– Next Generation Internet – Internet2
– GENI initiative
• Performance researcher should care about the end user
– What is the use of your studies?
– How can you relate your results to user perception?
References
1. A. Bouch, A. Kuchinsky, and N. Bhatti. Quality is in the eye of the beholder: Meeting user's requirements for Internet quality of service.
Technical Report HPL-2000-4, HP Laboratories Palo Alto, January 2000.
2. Nokia White Paper: Quality of Experience (QoE) of mobile services: Can it be measured and
improved?
http://www.nokia.com/NOKIA_COM_1/Operato rs/Downloads/Nokia_Services/whitepaper_qoe_
net.pdf
3. D. Soldani, M. Li, and R. Cuny, eds. QoS and QoE Management in UMTS Cellular Systems.
Wiley, 2006
References
4. M. Fiedler, ed.: EuroNGI Deliverable
D.WP.JRA.6.1.1. State-of-the-art with regards to user-perceived Quality of Service and quality feedback. May 2004.
http://eurongi.enst.fr/archive/127/JRA611.pdf 5. M. Fiedler, ed.: EuroNGI Deliverable
D.WP.JRA.6.1.3. Studies of quality feed-back mechanisms within EuroNGI. May 2005.
http://eurongi.enst.fr/archive/127/JRA613.pdf 6. T. Hoßfeld, A. Binzenhöfer, M. Fiedler, and K.
Tutschku: Measurement and Analysis of Skype VoIP Traffic in 3G UMTS Systems. Proc. of IPS- MoMe 2006, Salzburg, Austria, Feb. 2006, pp 52—61
CfP
• WP.IA.8.6: First EuroNGI Workshop on Socio- Economic Impacts of NGI
• DTU, Lyngby (Copenhagen), Danmark, Oct. 9—
10, 2006.
• http://eurongi06.com.dtu.dk/
• Still accepting contributions (extended abstracts)
Thank you for your interest ☺ Q & A
markus.fiedler@bth.se Skype: mfibth