Intends to analyze the challenges for low latency services and upcoming new data opportunities. Describes the Internet data service waves: Internet Mobile; Internet of Things and now Tactile Internet, how low latency is becoming the king and new data services, the main attribute. Tactile Internet will open immersive services experience beyond current audiovisual ones, but it depends to understand the human sense reaction idiosyncrasies. Also for IoT, how latency will affect overall productive processes since new robotics control for Industry 4.0 to remote surgery. It introduces the Edge Computing as one the most important toll to reduce latency by facilitating data processing at or near the source of data generation. I.e., Edge Computing avoids the application data travels from the source to very far application and processing hosts. But the latency reduction does not have an unique strategy, and it requires to have transport network proper dimensioned and configured can certainly improve the network low latency. In addition, new transport architecture with simplification collapsing network layers, network aware multidomain control, automation etc. will be definitively imperative for optimizing latency for new critical applications. Improving processor performance using technologies beyond regular x86 servers, such as: GPUs, FPGA, DPDK etc, can reduce the time of workloads and delivering/returning data to user application with readiness. In terms of low latency opportunities, accordingly to Chatan Sharma Consulting, Edge Internet economy will be over $4.1 Trillion worldwide. The initial growth will come from Edge serving existing use cases and will gradually be accelerated by the new use cases as the deployment becomes more widespread and developers learn to take advantage of the Edge Internet architecture for applications across industry domains in all major markets. But for MNOs, Edge Computing will open a several opportunities such as Service Integration, Managed Services, IaaS/PaaS/SaaS, new collocation business modelling etc. Besides latency, in general Edge Computing can bring an immediate benefit around local traffic offloading that implies in reducing overall CapEx cost. Higher Traffic Density can rapidly justify the Edge Computing deployment.

1. Network Challenges and New
Business Modeling for Low
Latency Services
September, 18 & 19th 2019
Alberto Boaventura
Ger. Estratégia e Arquitetura de Rede
Diretoria de Estratégia, Tecnologia e Arquitetura de Rede
Conecta LATAM - TELCO DIGITAL
TRANSFORMATION 2019

2. Mid 00’: “Every 100 ms of
delay costs 1% of sales”
Mid 00’: “Extra 0.5s in
search page generation
time dropped traffic by
20%.”

3. Telecommunication Industry Scenarios: Evolution Waves
Internet Mobile Internet Internet of Things Tactile Internet
Transformation
Enabled global communications;
New form of entertainment;
Communication cheaper;
Communications everywhere;
Reachability of users and services;
Smartphone: becoming the main device for
access to content;
Smartphones => Beginning of Digital
Transformation;
Potentialized the use of Social Networks;
It will revolutionize all industrial segments
through electronic integration and
transaction, improving and optimizing its
production processes;
New innovation platform,
inaugurating a new economic moment;
Experience not only audiovisual, but full
immersion;
They will help users (humanity) in
complementing the perception of the
world by bringing more information
through sophisticated applications of
Augmented Reality and Virtual Reality;
Market
Voice traffic superior to data;
Voice is the most important service;
Non-Real-Time Applications
Fixed broadband growth is 42% (CAGR)
against 22% for mobile accesses;
Fixed Broadband> 64 kbps;
Default ADSL.1 = 1-8 Mbps
Growth of fixed broadband is 8% (CAGR)
against 33% for mobile broadband;
The number of MBB exceeds that of FBB in
2013;
Data traffic already outperforms voice
traffic. Data becomes more the most
important service;
In the next decade will be some tens of
billions of connected objects Explosion of
connected objects;
Services with low latency and security have
another relevance;
The 5G is promising with: expectation of 1
billion subscribers in less than 5 years of
operation; above 10 trillion in revenue
over the same period;
Latency becomes more relevant than
broadband for tactile Internet services;
Technology
World wide web
Dial Up
xDSL
Packet switching and TCP / IP
WAP
EDGE, HSPA / HSPA + and LTE / LTE-A
Android
Video streaming and sharing
Social networks
Global Roaming
Access LPWA, SigFox, LoRA, LTE-M
Computing in the Cloud;
Cognitive computing;
Big Data, Analytics, Machine Learning, AI
Blockchain
Low latency systems: SDN / NFV;
Edge/Fog Computing
Quantum Computing
Quantum Internet;
2020201020001990
Broadband Mobility Connectivity Latency

4. Above 500 ms 100 ms to 500 ms 50 to 100 ms 5 to 50 ms below 5 ms
Human
Experience
IoT&Business
Requirements
ApplicationsSample
•Non real time services;
•Web browsing;
•Email;
•Non real time smartmeetering:
Utility Sensors;
•Retail better than 500 ms;
•Satellite voice services
•Video on demand streaming
•Cognitive assistance (20-40 msec)
•Virtual reality
•UAV control (10 – 50 msec)
•Automotive pre-crash sensing
warning
•Vehicle safety apps (mutual
awareness of vehicles for
warning/alerting)
•Assisted driving – cars make
cooperative decisions, but driver
stays in control
•Realtime smartmeetering:
eHealth Sensors
•Telepresence with real-time
synchronous haptic feedback
•Industrial robots; Closed-loop
industrial control systems
•Smart grid: Synchronous phasing
of power supplies
•Shared haptic virtual environment
•Tele-medical applications, e.g.
tele-diagnosis, tele-rehabilitation
•Augmented reality
•Cooperative driving (20 msec)
•Serious gaming (20 msec)
•Industrial Automation
•Remote Robots/Surgery
•Industry 4.0
•Negotiated automatic
cooperative driving
Why latency is so importante for upcoming services?
Body Reaction Audio Reaction Visual Reaction
Cortical Neuron
Communication
Remote Surgery
& Robotics
Industrial
RobotsUAV ControlRetailSmartmetering
Web Access

5. BBU
BBU
5G Latency dependence
TDD BANDS INTERFERENCE 5G URLLC & UPF ON THE BORDER
ICC & COMP MASSIVE MIMO
DL UL DL UL
Victim
Victim
BS to BS
Interferation
UE to UE
Interferation
Aggressor
Aggressor
PCRF
HLR/HSS
OCS/
OFCS
S-GW P-GWMME
IMS
RRHRRH
Internet
Oper. A Oper. B
500ms – 1s
1 ms
100µs
5µs
1 µs
150ns
65ns
LegacyNetworks(2G,3G,
4G)&Billing
5GURLLCRequirmeent
AssynchronousDual
Connectivity(100µs)
TDDBands
CoMP&ICIC
CAIntra-band
massiveMIMO
X 10-3
X 10-1
X 10-1
X 10-1
X 10-1
X 10-1
LogarithmScale
X2 (1 µs)
Victim Cell
hnm
h21
h12
h11
Internet
<1 ms
X2 (1 µs)
h11
h12 h21
h22
65 ns
±50 ppb
(Wide Area)
±100 ppb
(Medium/Local Area)
±250 ppb
(Home Area)
Master Clock or
GPS closer to
access
Latency & network synchronization is a key
enabler for 5G network and services: since
to provide new tactile internet experience till
to allow algorithms to resource and
interference management and optimization
for improve network capacity
<1 ms
5µs

6. Latency & Network Architecture/Topology Changes
UP AND CP SEGREGATION  USER EXPERIENCEC-RAN  CAPACITY IMPROVEMENTCURRENT LTE NETWORK
AtCoreAtEdge
Internet
Backhaul
...
PCRF
HLR/HSS
OCS/
OFCS
S-GW P-GWMME
IMS Internet
Backhaul
RANgoestotheCenter
InterferenceMitigation
IncreaseCapacity
Internet
CoregoestotheEdge:
Lowlatency
TrafficOptimization
Interference between cell
sites redeuces the overall
network capacity
ThedistancebetweenUEand
Internetandgreatnumberof
hopsproducesahugeimpact
inlatency
Backhaul
BBU
RRH
BBU
RRH
PCF
UDR
SMF
UPF
AMF
Other
RURU
Fronthaul
DU
DU
Midhaul
0-10 km
50-200 µs
100 G+ links
L2
P2P & P2MP
20-40 km
< 2ms
100 G+ links
L2/L3
40-200 km
< 10 ms
400 G+ links
L3
RURU
Fronthaul
DU
DU
Midhaul
0-10 km
50-200 µs
100 G+ links
L2
P2P & P2MP
20-40 km
< 2ms
100 G+ links
L2/L3
40-200 km
< 10 ms
400 G+ links
L3
PCF
UDM
SMFAMF
Other
CU
CU
UPFCU
CU
≤ 1ms
1µs
1µs

7. Reducing the Latency: From the Core to the border
TDC Distribution
PCF
SMFAMF
TDC Access
CU
CU
UPF
TDC Core
UDMNSSF
NEF NRF
SDN Controller
DU
DU
FronthaulMidhaulBackhaulInternet
10 to 2000 ms > 50 ms > 10 ms < 2ms < 0.20 ms 1 to 10 ms
Cloud Processing Internet Backhaul+Backbone Midhaul Fronthaul Radio Access
3rd Party Applications
3dr party DC
RU
RU
PoD
Backbone
Delay Budget
Cloud Computing Edge Computing
Usage
Characteristics
•Control Plane
•Core Applications and General Internet Users
•Global information collocated from worldwide
•Centralized, IT based services
•Faraway from users and communicate through IP networks
•User Plane
•Mobile users
•Limited localized information services related to specific deployment
locations
•Low latency dependency and telecom services
•In physical proximity and communicate usually through single-hop
wireless connection
Working &
Deployment
Environment
•Warehouse-size building with cooling and precision air
conditioning systems
•Ample and scalable storage space and compute power
•Centralized and very big buildings
•Outdoor (streets, parklands, etc.) or indoor (restaurants, shopping
malls etc.)
•Limited storage, compute power and wireless interface
•Centralized or distributed in regional areas by local business
(DLSAM/OLT Cabinets, RNC/BSC/BTS, local teleco operator or vendor,
shopping mall retailer etc.)
Competition •High competition with big players: Google, Microsoft, Amazon •Low competition => Competitive advantage to operators
3rd Party Applications

8. Simple Edging analysis
371
334
265 247
196 193 188 177
137 121
43
PA BA PR MG SC CE RS PE GO SP RJ
Most populous states and the dispersion index
(km)
𝑰𝒏𝒅𝒆𝒙 =
σ𝒊 𝑷𝒐𝒑𝒊 × 𝑮𝒆𝒐𝑫𝒊𝒔𝒕𝑖
σ𝑖 𝑃𝑜𝑝𝑖
IMPACT OF TWO NEW CLUSTERSIMPACT OF INCREASING EDGING IN BAHIAPOPULATION DISPERSION IN BRAZIL
81
7 4
54
SALVADOR ITABUNA TEIXEIRA DE
FREITAS
VITÓRIA DA
CONQUISTA
VITÓRIA DA
CONQUISTA
SALVADOR
ITABUNA
TEIXEIRA DE
FREITAS
Dispersion index (km)
33
7
4
18
12
25
SALVADOR ITABUNA TEIXEIRA DE
FREITAS
VITÓRIA DA
CONQUISTA
BARREIRAS JUAZEIRO
VITÓRIA DA
CONQUISTA
SALVADOR
ITABUNA
TEIXEIRA DE
FREITAS
Dispersion index (km)
BARREIRAS
JUAZEIRO
SALVADOR

9. Reducing the Latency: Transport Delay
Incoming Ramdom
Traffic
Outgoing Shapped
Traffic
Packet delayed due
queue length
No packet
delay

M/M/1
Router

𝑾 (𝒍𝒂𝒕𝒆𝒏𝒄𝒚) =
𝟏
𝝁 − 𝝀
W

W
Queueing delay
Average Incoming packet throughput Outgoing packet throughput
Propagation delay
Fiber Length - km (l)
Cladding
𝑷(𝒎𝒔) = 𝒍 × 𝟑. 𝟑𝟒
Outgoing port throughput delay
0 1 ... 1 0 1 0 0 1 ... 1 0 1 0
0 1 ... 1 0 1 0
Output
delay
𝑶 =
𝑩

Packet Length (B)
Processing (Routing/Switching) delay
Memory Bus Crossbar
𝑹 < 𝟏𝟎𝝁𝒔 𝒑𝒆𝒓 𝒏𝒆𝒕𝒘𝒐𝒓𝒌 𝒆𝒍𝒆𝒎𝒆𝒏𝒕
𝑻𝒓𝒂𝒏𝒔𝒑𝒐𝒓𝒕 𝑫𝒆𝒍𝒂𝒚 = 𝑵 × (𝑹 + 𝑶 + 𝑾) + 𝑷

10. UDM NSSF
NEFNRF
Reducing the Latency: Transport Optimization & Evolution
OTIMIZAÇÃOTRANSPORT NETWORK FOR 5GCURRENT TRANSPORT NETWORK
InnerAcesso Distrib.
DWDM OTN/
Core
OTN/
Acesso
BTS
Optical
NetworkIpNetwork
Mobile
Network
PCRF
HLR/HSS
OCS/
OFCS
S-GW P-GWMME
IMS
RU
TDC Access TDC Distribution
BTS
Spine-Leaf
PCF
SMFAMFCU
CU
UPF
RU
TDC Core
Multiple Hierarchies
Independent Optical and IP Domains
Distributed Control
Low scalability
Simplification
E2e Centralized and Multi-Domain Control
Virtualization
Elastic and Scalable
Open and Flexible Architecture
Automation
SDN Controller
Midhaul Metro
DWDM
OTN/
Core
Fronthaul
DUDU
IP/SRAcesso Distrib.
Rede TSN
IEEE 1588
Edge Computing Cloud Computing
Fronthaul Midhaul Backhaul Backbone

11. Reducing the Latency: ImpRove HW Architecture
CPU
Control Plane
Storage
Storage
Communication
Data Flow Prrocessing
IO
CPU
FPGA
FPGA
FPGA
GPU
Compute
Acceleration
FPGA
Control Plane
• Management
• Securrity
• Protocol Bridging
Storage
Storage Acceleration
• Cryptography
• Compression
• Indexing
Data Flow Prrocessing
• Inline processing
• Pre-processing
• Pre-filtering
• Exception processing
• Cryptography
• Compression
• IO expansion
• Protocol bridging IO
Ordinary x86 CPU Server Server Architecture with Accelerators
1X 1X 1X
4X
10X 10X
21X
27X
36X
Speech
Inference
Video
Inference
Recomender
Inference
CPU Server Tesla P4 Tesla T4
CPU
GPU
FPGA
ASIC
Flexibility
Cost
Efficiency
CPU GPU FPGA
• Low compute density
• Complex control logic
• Large caches
• Optimized for serial operations
• Shallow pipelines
• Low latency tolerance
• High compute density
• High computattions per memory access
• Built for parallel operations
• Deep pipelines
• High latency tolerance
• Field Programmable Gate Arrays
• (FPGAs) are based around a matrix of
configurable logic blocks (CLBs)
connected via programmable
interconnects.
• Can be reprogrammed to desired
application or functionality
• FPGAs uses less power, but new NVIDIA
chips use as few as 10-15 watts for a
teraflop.
• Verification of complex designs
implemented is worse than CUDA code.

12. Reducing the Latency: Choose the right SW Architecture
TDC Distribution
PCF
SMFAMF
TDC Access
CU
CU
UPF
TDC Core
UDMNSSF
NEF NRF
SDN Controller
DU
DU
FronthaulMidhaulBackhaulInternet
3rd Party Applications
3dr party DC
RU
RU
PoD
Backbone
Virtual Resources
Virtualization Layer
HW
VNF Instancies
VNF1 ... VNFn
Lifecycle Service Orchestrator e2e
EMS
Data Plane Development Kit (DPDK)
NFVI
Container Eng.
Host OS
HW
CNF Instancies
CNF1 ... CNFn
MANO
NFVO
CNFM
CIM VIM
VNFM
Virtual Resources
Virtualization Layer
HW
Mobile Edge
Platform
Data Plane Development Kit (DPDK)
NFVI
Container Eng.
Host OS
HW
Mobile
Applications
Mobile Edge Host
Level Managger
Mobile Edge
Platform Manager
CIM VIM
Mobile Edge
Orchestrator
User App. LCM
Proxy
OSS/BSS
EMS
OSS/BSS
At EdgeAt Cloud
ETSI NFV Architecture ETSI MEC Architecture

13. unlocking new opportunities with Low Latency services
Industry 4.0 Healthcare Entertainment Transport Other Industries
• Industrial Automation
• Digital twin
• AR and VR Applications
• Smart Factory/Industrial
• Automation
• Industrial Control
• Robot Control
• Process Control
• Manufacturing Industry
• Motion Control
• Remote Control
• Remote Diagnosis
• Emergency Response
• Remote Surgery
• Digital Twin
• AR, VR & Immersive
Entertainment;
• Virtual Tourism;
• Sports VR & 360 video
streaming;
• Online Gaming;
• Game Streaming;
• Sport Bettings;
• Driver Assistance Applications
• Automated & Autonomous
Driving;
• Enhanced Safety;
• Cooperative collision avoidance
(CoCA) of connected automated
vehicles;
• Video data sharing for assisted
and improved automated driving
(VaD);
• Changing driving-mode;
• Traffic Management;
• Airlines/Airports;
• Utilities: Smart Energy, Smart
Grid
• Critical Metering
• Financial Industry: Stock
transaction; Electronic Fund
Transfer
• Insurance Industry
• Smart Cities: Security; Urban
Mobility
• E-Commerce
According to Chetan Sharma Consulting - August 2019: “By 2030, the Edge Internet economy
(Low Latency Services) will be over $4.1 Trillion worldwide. The initial growth will come
from Edge serving existing use cases and will gradually be accelerated by the new use cases as the
deployment becomes more widespread and developers learn to take advantage of the Edge Internet
architecture for applications across industry domains in all major markets”.

14. Beyond low latency Services: Edge computing opportunities for MNOs
E2E Costumer
Applications
B2B2x Solutions System Integration Edge IaaS/
PaaS/NaaS
Dedicated Edge
Hosting
Description&
Scenario
Consists of offering a top latency
based services;
For instance, VR 360 content
streaming;
Consists of offering specific vertical
services;
For instance, operator offers an
edge-enabled caching/analytics
solution for CCTV cameras;
Operator offers a system
integration and managed services
to large enterprise customers;
Operator provides MEC-enabled
IaaS/PaaS/NaaS to a large number
of developers/enterprise customers;
Customers buy MEC-enabled
compute/storage/services etc.
through a self-service cloud portal;
they deploy (parts of) their
application;
Operator acts as a collocation
provider, edge server are
deployed on a as-needed basis
depending on the geographical
coverage request by the public
cloud provider;
Revenues
• One-time revenue: Initial
upshot payment per subscriber;
• Recurring revenues: Monthly
fees per subscriber;
• One-time revenue:
Configuration fees (charge per
camera/device connected);
• Recurring revenues: Service fees
(char per camera/device
connected);
• One-time revenue: Plan & build
fees;
• Recurring revenues: Manage
service and planned, ongoing
support; Additional feature
development (not planned)
• Recurring revenues: Pay-as-go
based on
compute/storage/connectivity
resource and tier;
• One-time revenue: Initial
preparation fees for
configuration/Installation of
hardware
• Recurring: Compute/storage
fees; Connectivity; APIs
usage-based fees
Costs
• One-time costs: Initial service
development; VR headsets;
• Recurring costs: Content rights;
Ongoing feature development;
• One-time costs: Server HW
(MEC) costs; Infra and
Installation of server;
• Recurring costs: Basic
operations (power, cooling,
collocation); Monitoring, Field
maintenance; troubleshooting;
SW licensing; Ongoing feature
development
• One-time costs: Plan & build
costs (HW & SW components,
development costs,
professional services fees)
• Recurring costs: Operating
costs (e.g. for MEC infra,
planned support etc.);
Additional feature development
costs;
• One-time costs: Server HW
(MEC) costs; Infra and
Installation of server;
• Recurring costs: Basic operations
(power, cooling, collocation);
Monitoring, Field maintenance;
troubleshooting;
• SW licencing;
• One-time costs: Server HW
(MEC) costs; Infra and
Installation of server;
• Recurring costs: Basic
operations (power, cooling,
collocation); Monitoring, Field
maintenance;
troubleshooting;
IaaS,
SaaS, PaaS
Source: HP & Intel – STL Partners: “EDGE COMPUTING: 5 VIABLE TELCO BUSINESS MODELS 2017”

15. Edge Data Center Infrastructure
Cloud Data Center Edge Data Center
Power • Purpose-built utility / site electrical feeds
• Multiple grid feed for redundancy
• Diesel generators provide long-term energy source during utility
outage
• May leverage existing power infrastructure repurposed for edge computing
• May lack redundant power grid feed
• Diesel generator backup may not be feasible due to space, noise, or pollution
restrictions
Cooling • Purpose-built
• Cooling capability and redundancy built by design
• Cooling design options may be limited due to the size or location of data
center
Connectivity • Redundant connectivity is typical
• Design for performance
• Redundant connectivity is desirable but depends on applications and site
restrictions
• Design for low latency
Facility • Purpose-built facility or room
• Dedicated or MTDC
• Flexibility in site location, to reduce hazards
• Typically manned, with varying levels of monitoring and automation
• Purpose-built facility, leverage existing facility, or purpose built enclosure »
Dedicated or MTDC
• Latency and location of compute needs drive EDC location
• More likely to be unmanned, with remote monitoring and automation
Source: TIA 2019

16. Low Latency Brief Capex Analysis for single application
...
...
CloudComputingEdgeComputing
N nodes
𝐶 𝑇
𝐶 𝑇
𝐶 𝑇
𝐶 𝑇
𝐶 𝑇
𝐶 𝑇
𝐶 𝐸𝐷𝐶
𝐶 𝐸𝐷𝐶
𝐶 𝐸𝐷𝐶
𝐶 𝐸𝐷𝐶
𝐶 𝐸𝐷𝐶
𝐶 𝐸𝐷𝐶
𝐶 𝐸𝐷𝐶
𝐶 𝐸𝐷𝐶
𝐶 𝐷𝐶
𝐶𝒍𝒐𝒖𝒅𝑪𝒐𝒔𝒕 = 𝑵 × 𝐶 𝑇 + 𝐶 𝐷𝐶
𝐶𝒍𝒐𝒖𝒅𝑪𝒐𝒔𝒕 = 𝑵 × 𝐶 𝑇 + 𝑀 × 𝐶𝑆 + 𝐶𝐼𝑛𝑓𝑟𝑎𝐷𝐶
𝑬𝒅𝒈𝒆𝑪𝒐𝒔𝒕 = 𝑵 × 𝐶 𝐸𝐷𝐶
𝐶 𝑇 = 𝑐 𝑇 × 𝑹 × 𝑻𝒓𝒂𝒇𝒇
𝑅
𝐶 𝑇 = 𝑐 𝑇 × 𝑹 × 𝑫𝒆𝒏𝒔 × 𝑲 × 𝑹 𝟐
= 𝑲 × 𝑫𝒆𝒏𝒔 × 𝑹 𝟑
𝐶𝒍𝒐𝒖𝒅𝑪𝒐𝒔𝒕 = 𝑲 × 𝑫𝒆𝒏𝒔 × 𝑹 𝟑
+ 𝑀 × 𝐶𝑆 + 𝐶𝐼𝐷𝐶
𝑬𝒅𝒈𝒆𝑪𝒐𝒔𝒕 ≈ 𝑀 × 𝐶𝑆 + 𝐶𝐼𝑛𝑓𝑟𝑎𝐸𝐷𝐶 + 𝜎 𝐸𝐷𝐶
𝐶𝒍𝒐𝒖𝒅𝑪𝒐𝒔𝒕 = 𝑲 × 𝑫𝒆𝒏𝒔 ×
𝑳
𝟑. 𝟑𝟒
𝟑
+ 𝑀 × 𝐶𝑆 + 𝐶𝐼𝐷𝐶
𝐶𝒍𝒐𝒖𝒅𝑪𝒐𝒔𝒕 − 𝑬𝒅𝒈𝒆 𝑪𝒐𝒔𝒕 =
𝑲 × 𝑫𝒆𝒏𝒔 ×
𝑳
𝟑. 𝟑𝟒
𝟑
+ 𝑀 × 𝐶𝐼𝐷𝐶 − 𝐶𝐼𝑛𝑓𝑟𝑎𝐸𝐷𝐶
− 𝜎 𝐸𝐷𝐶
𝐶𝒍𝒐𝒖𝒅𝑪𝒐𝒔𝒕 𝟏
𝐶𝒍𝒐𝒖𝒅𝑪𝒐𝒔𝒕 𝟐
𝑬𝒅𝒈𝒆𝑪𝒐𝒔𝒕
𝑫𝒆𝒏𝒔 𝟏 > 𝑫𝒆𝒏𝒔 𝟐
L (Latency)
Besides latency, in general
Edge Computing can bring an
immediate benefit around local
traffic offloading that implies in
reducing overall CapEx cost.
Higher Traffic Density can
rapidly justify the Edge
Computing deployment
Infrastructure
Cost

17. Low Latency Opex Analysis
Hyperscale DC
1 Big Place
TDC Core
1 - 20 Places
TDC Distribution
10-50 Places
TDC Access
50-100 Places
PODs
> 1000
Devices
Millions of
Devices> 5000 Servers 100- 500 Servers 50-100 Servers 1- 10 Servers 1 Server
1 GW 100 MW 10 MW 1 MW 100 kW 10 kW 1 kW 100 W 10 W 1 W 100 mW 10 mW
Source: Based on Disruptive Analysis 2018
Processing
Chip
(<0,2 mW)
RaspberyyPi
(0,5 mW)
Edge
Platform
(30 W)
Standard
Server
(5 kW)
TDC Access
(EdgeMicro)
(50 kW)
TDC
Distribution)
(500 kW)
TDC Core
(2 MW)
NVIDIA
Pegaus
(500 W)
Huperscale DC
(1 GW)
Hyperscale DC TDC Core TDC Distribution TDC Access PODs
Power
Consumption
> 100 MW 400 kW- 2 MW 100 kW- 500 kW 20 - 50 kW 1-5kW
Area > 10,000 m2 200-1,000 m2 100 - 200 m2 20-100 m2 4 m2
Number of Places 2x1 2x1-20 10-50 50-100 1000-10000
Total Power
Consumption
1 GW 1.6 - 40 MW 1 - 25MW 1 - 5 MW 5-50 MW
Total Area > 20,000 m2 400-20,000 m2 1,000-40,000 m2 1,000 - 5,000 m2 4,000-40,000 m2
Cell Site based OpExCore Network based OpEx
PCF
SMFAMF
TDC Access
CU
CU
UPF
TDC Core
UDMNSSF
NEF NRF
SDN Controller
DU
DU
FronthaulMidhaulBackhaulInternet
3rd Party Applications
3dr party DC
RU
RU
PoD
Backbone
TDC Distribution

18. Grupo A
Norte
Grupo B
Sul
Grupo C
Polígono
DF/MG/RJ/SP/PR
Grupo D
SE/NE
How Oi can fit new Low latency business
Oi is the operator that has the largest fiber optic network in the country with over 375,000 km of fiber cables for local and long
distance access;
The Company has a large long-distance fiber optic network serving more than 2000 Brazilian municipalities that are being upgraded
to support 100 Gbps speeds;
Oi has wide real estate facilities to support fixed and mobile networks, such as thousands of DSLAM & OLT cabinets, remote stages,
cell sites, buildings, etc.
All installation sites for TDCs follow criteria with double fiber approach - OTN Network Sites with reliable transmission approach;
Preference for sites with InnerCore and Outercore routers; Lower latency between sites; Better traffic distribution among regionals;
NETWORK OPTICAL TRANSPORT TOPOLOGY FIXED BROADBAND OFFERING TDC CORE WITH HIGH TRANSPORT AVAILABILITY
Optical Network & Facilities
99%
89%
47%
98%98%
73%
23%
96%93%
28%
1%
80%
Oi Vivo TIM Claro
PIB População Município
Fonte: : Anatel/IBGE 2018 Fonte: : GTV Plano de Virtualização 2017/2018

19. Summary
● Low Latency is becoming the king: The new wave for upcoming data services will have latency as main attribute. Low latency will allow inedited immersive/real time
services, becoming a new internet era: Tactile Internet. There are lot of technologies that enable low latency and catalyze Tactile Internet, such as: 5G,& Edge
Computing;
● Tactile Internet requirements: The immersive services experience depends on which type of human sense reaction is used, and vision reaction is one critical one,
which requires latency below 5 ms. New robotics control for Industry 4.0, will require a very low latency as well for proper production integration;
● Low Latency is a key enabler for 5G: Since to provide new Ultra Reliable Low Latency Communication Services till to allow algorithms to resource and interference
management and optimization for improve network capacity
● Edge Computing: It is the most important toll to reduce latency and consists to facilitate data processing at or near the source of data generation (Gartner). Thus, it
avoids the application data travels from the source to very far application and processing hosts;
● Reducing the Latency. The overall transport network proper dimensioned and configured can certainly improve the network low latency. However, new transport
architecture with simplification collapsing network layers, network aware multidomain control, automation etc. will be definitively key aspects for optimizing latency for
new critical applications. Improving processor performance using technologies beyond regular x86 servers, such as: GPUs, FPGA, DPDK etc, can reduce the time of
workloads and delivering/returning data to user application with readiness;
● Upcoming Low Latency Opportunities: Edge Internet economy (Low Latency Services) will be over $4.1 Trillion worldwide. The initial growth will come from Edge
serving existing use cases and will gradually be accelerated by the new use cases as the deployment becomes more widespread and developers learn to take
advantage of the Edge Internet architecture for applications across industry domains in all major markets;
● Edge Computing Opportunities: The opportunities for MNOs will be beyond of supporting directly services and use cases to end customers. It can open a several
opportunities such as Service Integration, Managed Services, IaaS/PaaS/SaaS, new collocation business modelling etc.
● Capex Optimization: Besides latency, in general Edge Computing can bring an immediate benefit around local traffic offloading that implies in reducing overall CapEx
cost. Higher Traffic Density can rapidly justify the Edge Computing deployment
● How Oi can help; Oi is the operator that has the largest fiber optic network in the country with over 375,000 km of fiber cables for local and long distance access; It has
a large long-distance fiber optic network serving more than 2000 Brazilian municipalities that are being upgraded to support 100 Gbps speeds; It has wide facilities to
support fixed and mobile networks, such as thousands of DSLAM & OLT cabinets, remote stages, cell sites, buildings, served by abundant optical transport network.

20. ¡Gracias!
Thanks!
Obrigado!
Q&A
alberto@oi.net.br