Post on 14-Feb-2018
“Bit-Rate” and “Application Performance” in Ultra BroadBand Networks
4ºFocus: Gianfranco
Ciccarella - Telecom
Gianfranco Ciccarella - Telecom Italia Vice President Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
Telecom Italia
Index
► QoE platforms: the reason why
► How to improve Quality of Experience
► New business models
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
cv
cv
cv
IP traffic growth
247
46 23 72 80
865
134
62
259
320
World Western Europe LATAM North America APAC
2012 2017
x 3,5
x 2,9 x 2,7
x 3,6 x 4,0
Peak Internet Traffic – Total Bandwidth (Terabit/s)
Internet Video Traffic (% of total consumer traffic)
57% 51%
57%
66%
55%
69% 71% 76%
73%
66%
World WesternEurope
LATAM North America APAC
2012 2017
Source: Cisco VNI; Analysys Mason
1,2 1,3
3,2
4
2
9
Italy Argentina Brazil
2012 2017
x 3,4 x 1,5
x 2,8
Average Internet – Total Bandwidth (Terabit/s)
Data traffic «volumes» and «bandwidth» are growing…
… and most of the IP traffic is Video
3
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
Main performance requirements
Business requirements on Download time
Strangeloop – content delivery summit 2012
Examples of service throughput requirements
2 CH
b’cast
Key performance Drivers:
• Downstream Application
throughput
• Download time
4
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
“Application Throughput” and “Bit-Rate”
Central
Office User
Premises
Broadband
Line
Transmission Bit-Rate
Max TCP-IP data
transfer Bit-Rate
WEB FTP Application
Throughput
16 Mbps*
* Measures at my home, Italy, feb 2014
13 Mbps*
7 Mbps* e.g. Large File
Download (with
a browser)
e.g. PC-web site
speed test Tool
e.g. ADSL BB-
line monitor
• Application Throughput is
lower than Bit Rate (in some
case much lower!)
• High Application Throughput
requires low delay and low
packet loss
• QoS functionalities cannot
improve Throughput
• QoE Platforms (Application &
Content Delivery, WEB Acceleration,
Protocol Optimizations…) are needed
to improve best-effort IP
network performance and to
reduce network TCO
Application Throughput is the most important KPI affecting end-user Quality of Experience (QoE)
Web site
5
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
Application Throughput is lower than connection Bit Rate
Throughput/Bit-rate= 74%
6
Akamai average connection
speed ( throughput)
= 5,2 Mbps (4Q 2013)
The Average Connection Speed (a throughput proxy) metric represents an average of the measured connection speeds across all of the unique IP addresses seen by Akamai for a particular geography
Akamai Status of the Internet 4Q13 Report
Ookla SpeedTest ( bit rate)
= 6,98 Mbps (4Q 2013)
Download Speed Test (a bit rate proxy) uses up to four HTTP threads to saturate the user connection; several measure samples are analyzed to estimate the maximum connection speed.
Apr Jul Oct 2010 Apr Jul Oct 2011 Apr Jul Oct 2012 Apr Jul Oct 2013 2014
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
Application Throughput is lower than Connection Bit Rate… …and it gets worse for higher Bit Rates (1/2)
(*) Ookla SpeedTest data ( Bit Rate) - (3Q13) (#) Akamai connection speed ( Throughput ) - (3Q13)
For growing bit-rate: • the ratio Throughput / BitRate decreases • the “bit rate capacity waste” is more severe
7
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
Application Throughput is lower than Connection Bit Rate… …and it gets worse for higher Bit Rates (2/2)
(*) Ookla SpeedTest data ( Bit Rate) - (4Q13) (#) Akamai connection speed ( Throughput ) - (3Q13)
For growing bit-rate: • the ratio Throughput / BitRate decreases • the “bit rate capacity waste” is more severe
(4Q13)
8
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
Th
rou
gh
pu
t a
nd
Bit
Rate
[ M
bps ]
MAX TCP Throughput
Bit Rate and Throughput vs RTT and Packet Loss
(*) NetIndex/Ookla SpeedTest (4Q13) EU Commission report (oct.2013) shows that in EU:
P.Loss = 0,2% - 0,5% “Latency” = 19ms - 36ms
Rif: ● M.Mathis et Al. , “Macroscopic Behavior of TCP Congestion Avoidance Algorithm, July 1997
● Basso, et Al. Packet Loss Rate in the Access Through Application-Level Measurements, ACM SIGCOMM Helsinki, Aug 2012
TCP throughput over
xDSL/FTTx access
50 20 80 100
20
7
(**) Akamai State of the Internet 4Q13
Average DS
Throughput EU (**)
Average DS
Throughput Italy (**)
Average DS
BitRate EU (#)
(#) EU Commission report “Quality of BB in EU” (oct. 2013)
Long-term target
9
Average DS
BitRate Italy (*)
Telecom Italia
Index
► QoE platforms: the reason why
► How to improve Quality of Experience
► New business models
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
Content & Application Delivery networks, caching…
Reduce latency for better performance
WEB acceleration, Front End Optimization…
How to improve Quality of Experience
Network protocols optimization…
Multiple copies of contents
QoS functionalities cannot improve throughput nor reduce WEB page download time Telcos can leverage on both QoE platforms & QoS functionalities … and QoE platforms enable network TCO saving
Reduce user-server distance…
Quality of Experience (QoE)
• Subjective measure, from the user’s perspective, of the overall quality of the service provided.
• Usually expressed as “MOS, Mean Opinion Score”, ranging from 1 to 5.
• QoE is improved by platforms such as content & application delivery, caching, protocol optimization, front-end optimization, compression, adaptive bitrate.
Quality of Service (Network Level)
• QoS functionalities are always used in IP networks and provide traffic management mechanisms (e.g. IETF Diffserv: Differentiated Services, ...) based on different priorities. QoS is also needed in case of network congestion
Improve application
throughput
Reduce web page
download time
QoE
platforms
11
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
CDN/TC Platforms improve best effort network performance
Mobile Access
Fixed Access
All-IP Domestic Network
International
Network Core Metro Regional
UBB access
Network QoE Platforms. Policy Control.
Content nearer to users better performance (higher throughput)
Akamai 2012, «Empirical Network Analysis» 12
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
Throughput improvement by “protocol enhancer”…
From 2 to 5 times higher
throughput for many
traffic types (e.g. FTP,
HTTP, video HD…)
13
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
Web page download time improvement …
Keep Alive & Compression
Add a CDN Full optimization
1 2
3 4
QoE can be improved with a mix of technologies and solutions
Strangeloopnet “ WEB performance Automation” http://www.youtube.com/watch?v=IPn0T1UacIA
No web acceleration
sec
14
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
OTT are looking to “enter” in the Telco Networks
1
2
3
4
• QoE platforms managed
by OTTs
• Options 3 & 4 require IP
EDGE distribution
PoP
PoP
PoP
PoP
15
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
Network transformation guidelines
APP Server/ QoE Platf.
IP EDGE
MPLS
IP - Ethernet
ROADM WDM
MPLS
ROADM
IP - Ethernet
IP - Ethernet
Tx Network: ROADM in Core and Metro, WDM in Aggregation
End-to-end IP/MPLS on WDM for IP EDGE, Application Servers and QoE platforms distribution
QoE Platform deployment: Content Delivery Network Transparent caching Application Delivery Network Web Acceleration … …
APP Server/ QoE Platf.
AS-IS TO-BE
IP EDGE
APP Server/ QoE Platf.
IP EDGE
APP Server/ QoE Platf.
Key Points
16
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
20 ms < RTT < 25 ms (95%)
30 ms < RTT < 35 ms (95%)
RTT > 35 ms
Cache in the Net (1/2): RTT reduction
On Net
Caching in the Core
core sites
Off Net
Caching
Out of Telco ISP domain sites
On Net
Caching in the Metro
CO/metro sites
Internet
Interconnection
Point
Last Mile
BB-UBB access
areas
User’s
premises
17
Simulation Results RTT normalized distribution with
Caches deployed:
• at the interconnection point
(Off Net)
• in Core sites (On Net)
• in Metro sites (On Net)
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
On Net Caching in the Core core sites
Off Net Caching Out of Telco ISP domain sites
On Net Caching in the Metro CO/metro sites
Last Mile BB-UBB access
areas
User’s
premises
Cache in the Net (2/2): TCP throughput
18
Simulation Results Max TCP throughput distribution
with Caches deployed:
• at the interconnection point
(Off Net)
• in Core sites (On Net)
• in Metro sites (On Net)
PLR: 0,01%
PLR: 0,1%
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
Costs Saving due to Caches deployed in the network
19
Network Costs Saving (%) as a function of “Cx” (*)
(*) Cx =
“Upstream network “ cost = cost from the cache insertion point to the Big Internet interconnection
EC (Cache efficiency or HitRatio) = average % of traffic delivered by the cache
“Cache Fan Out” = traffic delivered by the cache (given by: EC * Traffic delivered to End Users
“downstream” the cache insertion point)
Unitary “upstream network” Cost [K€/Gbps]
Unitary Cache Cost [K€/Gbps] referred to “cache fan-out”
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
Costs saving due to distributed IP Edge & Caching Case Study
• IP services edge
• Peer caches
• Access aggregation
20 Tier 1 COs
99 Tier 2 COs
C = centralized architecture
D = distributed architecture
Network transport costs reduced 47%
A Bell Labs case study compared the TCO of a
centralized IP edge to a distributed IP edge with CDN
content caches over the five-year period. Network model
was based on a large Tier 1 service provider in NA
Source: ALU WP - VIDEO SHAKES UP THE IP EDGE; 2012
20
Telecom Italia
Index
► QoE platforms: the reason why
► How to improve Quality of Experience
► New business models
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
New business models
End-user
TELCO OTT Internet Access fees + premium services
CDN, ADN, ...
€ €
€ Services fees
Two–sides Business Model
Incremental revenues from OTT/CP: “Two-sides business model”
Telco Premium services offered to end users
To complement best effort IP traffic termination, Telcos are deploying QoE platforms and are offering differentiated quality for IP delivery to end-users and OTT/CP Examples of OTT/CP and Telcos agreements: Comcast/Netflix, Verizon/Google, Orange/Cogent, several Akamai agreements (including Akamai/Telefonica…)
Best-effort is not sufficient to meet requirements for Application/Content/Services in the ALL-IP scenario QoE platforms in the Domestic Network are needed
22
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
UBB+QoE = ACCESS MONETIZATION…
QoE enables Telco Premium services
QoE improvement enables incremental revenues from NGAN & LTE
Average UBB price uplift
Western EU
DSL Cable FTTx
To get a Premium Access Fee from UBB access:
• access bit-rate improvement is not sufficient
• higher application throughput and lower download time are needed
23
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
• Comcast must terminate Netflix traffic, to avoid end-users complaint & churn (~30% of Comcast customers are also Netflix customers)
• Comcast gets a “traffic termination“ revenue from Netflix, additional to the “access revenue” from end-users, in a two-sides business model. The Telco revenue is related to the “value” of the Application/Content that OTT/CP offers to end-users. In the Netflix case, the “value” is relatively small (“US $7.99 all you can watch”); in other cases, the “traffic value” is higher (e.g. Amazon services)
• Comcast goal is to handle Netflix traffic efficiently (…a huge traffic, e.g. 3Tbps, that will further grow)
• The agreement with Netflix facilitates Comcast use of more effective&efficient solutions to handle Netflix traffic (QoE platforms) enabling network cost saving
Netflix-Comcast case
24
NO deal • Payback Time = never
• Additional Revenues from OTT/CP • Payback Time estimate > 25 years
Deal, but no use
of QoE platforms
• Additional Revenues from OTT/CP • Capex saving = 25%-40% • Payback Time estimate = 10-15 years
Deal, and use of
QoE platforms
3 options
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
Internet Best Effort QoE capable network
Dumb Pipe
OTT
CP
Є
Є Access fee ADV
Є Differentiated quality
for IP delivery
OTT
CP
Є
Є
ADV
Є
Access fee
BestEffort Є
Content, Services
Content, Services
Є
Business models transformation guidelines
: Router/Server/Cache OTT: Over The Top; CP: Content Provider; ADV: Advertising
Premium Access Fee
Content, Services
New IP Interconnection Policy
Regulatory issues on Net Neutrality
25
Gianfranco Ciccarella – Telecom Italia – VP Global Advisory Services
10 º Fiberness – Pieve Santo Stefano (Ar) – Giugno 2014
26