2. Dr Neil Davies
Co-founder, Predictable Network Solutions Ltd
Ex: University of Bristol (23 years).
Former technical head of joint university/research institute (SRF/PACT).
Peter Thompson
CTO, Predictable Network Solutions Ltd
Ex: GoS Networks, U4EA, SGS-Thomson, INMOS & Universities of Bristol,
Warwick and Cambridge.
Authority on technical and commercial issues of converged networking.
Martin Geddes
Founder, Martin Geddes Consulting Ltd
Ex: BT, Telco 2.0, Sprint, Oracle, Oxford University.
Thought leader on future of telecommunications industry.
PREDICTABLE
NETWORK
SOLUTIONS
3. PREDICTABLE
NETWORK
SOLUTIONS
Dr Neil Davies
Co-founder, Predictable Network Solutionsex-ante network performance
The only Ltd
Ex: University of Bristol (23 years). engineering company in the world.
Former technical head of joint university/research institute (SRF/PACT).
• New mathematical performance techniques.
• Performance assessment methodology.
Peter Thompson • World’s first network contention management
CTO, Predictable Network Solutionssolution.
Ltd
Ex: GoS Networks, U4EA, SGS-Thomson, INMOS & Universities of Bristol,
Warwick and Cambridge.
Authority on technical and commercial issues of converged networking.
Consultancy on the future of telecoms and
Martin Geddes the Internet.
Founder, Martin Geddes ConsultingBusiness model innovation.
•
Ltd
Ex: BT, Telco 2.0, Sprint, Oracle, Oxford University.product ideation.
• Technology &
• Organisation industry.
Thought leader on future of telecommunicationsdevelopment.
• Public & private workshops.
21. Every application
has performance
needs. QoE Aspirations
time to first frame, contiguous playback
audio/video quality, lack of glitches,
interactivity (text, screen share)
Network Expectations
These always
loss ? delay
map onto a
statistical
bound on loss
and delay.
22. A Recent Bad Skype Experience
We did a 3-way
Skype call to
1.8M/448k ADSL – wholesale 20CN
demonstrate
the fitness-for-
Loss: 0.1%. purpose issue.
Delay: 40ms-50ms
We measured
the network
loss and delay
characteristics.
20M/2M Cable broadband 10M/1M ADSL Business LLU
Loss: <0.5%. Loss: Wandering from a typical 0%-2% up
Delay (one-way): 50ms-60ms, jumping to as high as 48% for a second or two.
to 500ms for a second or two, then back Delay: 50-70ms
23. Different speeds & characteristics
SLOW & STATIONARY LOSS/DELAY
Good Experience
Bad Experience
VERY FAST FAST
& VARIABLE DELAY & VARIABLE LOSS
24. Summary:
Speed is not the only difference
Bad Skype QoE
VARIABILITY
HIGH
LOW
Another
factor was SLOW FAST
affecting SPEED
outcomes.
25. Cable Broadband:
High speed, just as advertised
Cable user happens
to be on SamKnows
network measuring
service.
Figures show the
average throughput
is as claimed.
27. Averages hide essential detail
Average Delay (by Day) Average Delay (by Hour)
100 800
86.62 720.28
Reported by SamKnows
Reported by SamKnows
80 668.74
600
Delay (ms)
Delay (ms)
60
42.63
400
40
20 21.68 200 169.71
13.79 14.51 14.49 133.06
0 0
07-May 09-May 11-May 13-May 15-May 00:00 04:00 08:00 12:00 16:00 20:00 00:00
Delay varying too Day 9th May
fast for Skype to
Delay (during call – afternoon of 9th)
Directly measured
compensate
600
Delay (ms)
400
200
0
0 10 20 30 40 50 60 70 80
Per second during call
28. Quality of Experience:
Depends on multiple characteristics
Determines the chance
that applications can Most stationary connection had
adapt to the network the BEST Skype experience
STATIONARITY
HIGH
Fastest connection had the
WORST Skype experience
LOW
Stationarity is the SLOW FAST
SPEED
Determines which
inverse of
applications are feasible
variability
29. What does
this network
property
enable?
STATIONARITY
30. The application
Hierarchy of Need
We need to evolve our metrics to capture a
richer set of capabilities that contribute to
fitness-for-purpose.
3. Reasonable bounds on loss and delay
2. Sufficient stationarity
1. Sufficient capacity
Note: exact requirements are application-dependent
31. So what?
• Measurement is de-facto regulation
– If your measurements are partial, your regulation is
partial, and users receive a partial service!
• Need a market framework for
quantity and quality
– Measure all network properties that impact
application QoE
– Encourage broadband service providers to offer a
variety of trades between peak bandwidth and
stationarity.
32. What next for regulators?
• Internal education
– Dispel the bandwidth myth
– Raise capability to reason on network quality management
• Measure and report stationarity
– Allow service providers to compete on more than just (peak)
bandwidth
• Engage stakeholders in a dialogue on quality
– Raise external awareness and provide forum for collective industry
progress