Copyright © 2018 Aricent. All rights reserved.
Accelerating Edge Computing
Adoption
A developer and application centric architecture
for multi-access edge computing
Shamik Mishra
Assistant Vice President
Technology & Innovation
Aricent (Altran Group)
Out of the Box Network Developers Meetup
Santa Clara
4-Sep-2018
2
1. Drivers for Multi-Access Edge Compute
2. Questions from App Developers
3. The Complexity of Edge Compute for an Application Developer
4. Developer Experience & Edge Marketplace
5. Distributed Edge and 5G Platform
6. Some problems with Edge Storage
7. Compute Offload through Serverless Architectures
8. Aricent Edge Compute Reference Architecture
CONTENTS
3Copyright © 2018 Aricent. All rights reserved.
Drivers for Multi-Access Edge Compute
Vehicle-to-Vehicle communication, Multi-player A/R gaming, Immersive video experience in entertainment, remote surgery, robotics, etc. are
some of the applications which needs to offload computing to a “neighborhood” data center and get an ultra-low latency connectivity
Access Agnostic Connectivity
to Network Edge
Cloud Native Platform for Virtual
Network Functions and Applications
Low Latency Sensitive Application
to deliver improved User Experience
Billions of
Things &
Devices
OrchestrationCloud Native
Toolkits
Intelligent
Platforms for
Data & Context
High
Throughput
Offload
Computing to
Cloud
Edge
ApplicationsLow LatencyMulti- Access
Devices
Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
4Copyright © 2018 Aricent. All rights reserved.
Some Questions from Edge Application Developers
• Developer Experience - enabling developers by abstracting out the CSP network complexity
• For an average developer today, they run 10-12, maybe 20-40 workloads to serve end mobile users... on public cloud. They can easily manage and
monitor their applications as it is located in public cloud
• When this comes to edge, the developer will have to run hundreds and thousands of such workloads across data centres, across the country, or
beyond
• The overall cost per workload is going to be very high for the developer as well as for the operator.
• New Architectures with 1:1 device – server with additional support for messaging, notification, multi-tenancy, subpub would be required from the
platform.
• Go the cloud native way
• New kind of low latency workloads will require hardware accelerators like GPU, FPGAs etc.
• How much of these accelerators are ready for prime time, telco cloud usage. GPU is expensive, have limitations in virtualization
• What is the right edge for a client device? There must be a way to discover the right edge, based on location, latency, hardware capabilities, capacity
etc.
• In 5G / Edge Compute, there will be huge uplink data as more and more devices will start offloading computing to edge
• Need to find efficient ways of offloading compute to edge. Serverless / functions-as-a-service from device to edge could be one such solution
• What happens if an edge goes down? How to seamlessly move applications between device, edge and cloud...?
• Federating computing platforms across the device to edge to cloud… is this the answer?
Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
5Copyright © 2018 Aricent. All rights reserved.
The Edge Computing Complexity for an Application Developer
Aricent Confidential
The application
developer has no access
to the edge. It is deep
inside the service
provider network
Potentially hundred of application
instances running across several
edges. Monitoring and monetizing
for a developer is a nightmare
The telco network is too complex and
needs abstraction. NFV model of
application deployment may be too
complex for application developers
Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
6Copyright © 2018 Aricent. All rights reserved.
The Developer Experience and a Market Place
5G & Edge Market Place-as-a-Service
Onboarding
Service
(Developer
onboarding
portal)
Developer Services
(MEC Developer SDK, Developer Portal)
MEC Infrastructure Services
Business
Model
Realization
Marketplace-as-a-service enables operators to onboard developers by providing seamless developer experience (DX)
5G/Edge will not be cheap. Software Architecture needs to provide models to rationalize developer’s cost to develop and host
applications. Applications can be developed using pay-as-go models like functions-as-a-service / serverless / event based
architectures. This means developers pay only when it is used and does not host the application 24*7*365.
Edge Data
Center
Edge Data
Center
Edge Data
Center
Edge Data
Center
Edge Data
Center
Edge Data
Center
Edge Data
Center
Operator & Developer
Monetization
Application Developers
Application Developers
(applications, functions)
Micro-service Developers
(micro-services)
3rd Party App Provider
(applications, functions)
Distributed Edge Platform #1 Distributed Edge Platform #2
Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
7Copyright © 2018 Aricent. All rights reserved.
Distributed Edge and 5G Platform | Edge Compute
Ultra-reliable, low-latency high speed communications in 5G is expected to enable mobile digital solutions in healthcare, autonomous
cars, advanced robotics use-cases, augmented and virtual reality (AR/VR) applications and accelerates the already rapid growth of the
Internet of Things (IoT).
Data Center
eNB
Data Center
SGW
Data Center
PGW App
Data Center
eNB
Data Center
SGW PGW
App
Data Center
eNB SGW PGW App
L1
L2
L3
App
Massively Distributed Platform with
Horizontal Scaling and Hierarchical
Offloads
Source: Shamik Mishra, Vishal Gupta, Raveen Sharma.
Decomposable Distributed Architectures for low latency
sensitive applications. 2018. (Under publication)
Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
8Copyright © 2018 Aricent. All rights reserved.
Accelerated and Resilient Distributed Edge Storage
3,6
8,P
2,5
8,9
1,5
6,7
DATA
1
2
3
4
5
6
7
8
9
P
Generate Parity
4,7
9,P
1, 2
3,4
• Improvement in storage efficiency, Latency
• Employ new generation Clay Code.
• Clay Code has
• Least possible storage overhead
• Least possible repair bandwidth and disk read
• Shown 3x repair time reduction and up to 30%
and 106% improvement in degraded read and
write with CEPH
• Acceleration of Erasure Coding Offloading the
computation to GPU
• Integrate the accelerated erasure code algorithms
to CEPH
Accelerated Storage to ensure reconstruction and resilience of data over highly distributed edge. Data stored in encoded and
distributed blocks and can be reconstructed even if some edge is not accessible for a highly mobile client
Source: Dinesh Bhaskaran, Aricent. Accelerated Erasure Coding: The New Frontier of Software-Defined
Storage. 2018 Storage Developer Conference
Edge: For Caching / CDN / Edge Storage use cases, data can be stored in different geo-locations. Reconstructing data even
when certain edges are unavailable is critical to the overall performance.
Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
9Copyright © 2018 Aricent. All rights reserved.
Model Driven Distributed Applications
Hardware
Container Cluster
Service Mesh for Micro-Services (Scalable, Secure)
fn1 fn3fn2
Time
fn1
fn2
fn3 Low Latency -> edge
Cloud
DeviceDistributed App Composer (Model Driven Application Composition)
App Model App Configuration
fn1 fn2 fn3 AppDistributed
Deployment
fn1 fn2 fn3
FaaS
Event
Local
Diagnostics
Ensconce Management
Node
Dashboard
Cloud
Edge
Goals for the Approach
• Model drive application distribution and function composition
• Approach for Stateful functions
• Service Mesh for local diagnostics and “finding the needle in the haystack”
• Event driven architecture to define computing offload
• Auto-Configuration, scale-to-zero
• Infrastructure agnostic
• Distributed AI / Data Pipelines on container based infrastructure
Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
10Copyright © 2018 Aricent. All rights reserved.
Aricent Multi-Access Edge Computing Framework Reference Architecture
Edge Host
Edge Network
Function Platform
Edge Application
Platform
Platform & Application
Management
Edge
App
Edge
App
VNFs
Cloud IaaS
OSS / BSS
Application
Orchestration
Multi-Edge Management
5G and Edge Market Place as a service (Developer Experience DX)
Cloud
App
Cloud
App
Cloud
App
Cloud
App
Cloud
App
Cloud
App
ENSCONCE
Distributed Edge Platform
NetAnticipate
(Intent Based
Networking)
Client
App
Low Latency
Access to Apps
Distributed
Edge Host
Management
Orchestration
Policies
Billing
Application
Control
Platform
Management
MANO Platform
(Network Service
Orchestration)
Public Cloud
Operator Cloud
Operator
Edge
Network Automation and
Orchestration
Programmable Network Fabric
EMLEE
(Edge Machine
Learning Enhanced
Environment)
Integrated VNF
Frameworks
(Cloud-RAN, vEPC)
Edge PaaS
Edge Application SDK
DevOps and Platform
Bootstrap
Multi-Host
Orchestration and
Edge Discovery
Cloud Native
Application Platform
Serverless and
compute offload
Aricent Open Source Projects for Edge
Provisioning
https://github.com/cablelabs/snaps-boot
https://github.com/cablelabs/snaps-openstack
https://github.com/cablelabs/snaps-kubernetes
Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
11Copyright © 2018 Aricent. All rights reserved.
Key Objectives of Aricent Edge Compute Reference Architecture
Developer centric architecture for edge
applications with edge application SDK
Distributed and Event based Computing
Architecture to offload computing from
devices to edge through Serverless and
Sidecar design patterns
for edge applications with edge
application SDK
Multi-Edge Management mechanism
enabling devices (and developers) to
discover the “right” edge while executing
applications
Orchestration system to manage the
lifecycle of highly available network
functions and edge applications with
dynamic recovery, application migration
& security
Automated Bootstrap Complete
automated bootstrapping of micro-data
centers, platform-as-a-service and
management systems with minimal user
intervention
Accelerated Compute and Storage GPU
acceleration for low-latency compute.
Distributed edge storage through GPU
accelerated erasure codes
Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
12Copyright © 2018 Aricent. All rights reserved.
Thank You
Headquarters
3979 Freedom Circle
Santa Clara, CA 95054
www.aricent.com
Copyright © 2018 Aricent. All rights reserved.
Shamik Mishra
AVP – Technology & Innovation

Accelerating Edge Computing Adoption

  • 1.
    Copyright © 2018Aricent. All rights reserved. Accelerating Edge Computing Adoption A developer and application centric architecture for multi-access edge computing Shamik Mishra Assistant Vice President Technology & Innovation Aricent (Altran Group) Out of the Box Network Developers Meetup Santa Clara 4-Sep-2018
  • 2.
    2 1. Drivers forMulti-Access Edge Compute 2. Questions from App Developers 3. The Complexity of Edge Compute for an Application Developer 4. Developer Experience & Edge Marketplace 5. Distributed Edge and 5G Platform 6. Some problems with Edge Storage 7. Compute Offload through Serverless Architectures 8. Aricent Edge Compute Reference Architecture CONTENTS
  • 3.
    3Copyright © 2018Aricent. All rights reserved. Drivers for Multi-Access Edge Compute Vehicle-to-Vehicle communication, Multi-player A/R gaming, Immersive video experience in entertainment, remote surgery, robotics, etc. are some of the applications which needs to offload computing to a “neighborhood” data center and get an ultra-low latency connectivity Access Agnostic Connectivity to Network Edge Cloud Native Platform for Virtual Network Functions and Applications Low Latency Sensitive Application to deliver improved User Experience Billions of Things & Devices OrchestrationCloud Native Toolkits Intelligent Platforms for Data & Context High Throughput Offload Computing to Cloud Edge ApplicationsLow LatencyMulti- Access Devices Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
  • 4.
    4Copyright © 2018Aricent. All rights reserved. Some Questions from Edge Application Developers • Developer Experience - enabling developers by abstracting out the CSP network complexity • For an average developer today, they run 10-12, maybe 20-40 workloads to serve end mobile users... on public cloud. They can easily manage and monitor their applications as it is located in public cloud • When this comes to edge, the developer will have to run hundreds and thousands of such workloads across data centres, across the country, or beyond • The overall cost per workload is going to be very high for the developer as well as for the operator. • New Architectures with 1:1 device – server with additional support for messaging, notification, multi-tenancy, subpub would be required from the platform. • Go the cloud native way • New kind of low latency workloads will require hardware accelerators like GPU, FPGAs etc. • How much of these accelerators are ready for prime time, telco cloud usage. GPU is expensive, have limitations in virtualization • What is the right edge for a client device? There must be a way to discover the right edge, based on location, latency, hardware capabilities, capacity etc. • In 5G / Edge Compute, there will be huge uplink data as more and more devices will start offloading computing to edge • Need to find efficient ways of offloading compute to edge. Serverless / functions-as-a-service from device to edge could be one such solution • What happens if an edge goes down? How to seamlessly move applications between device, edge and cloud...? • Federating computing platforms across the device to edge to cloud… is this the answer? Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
  • 5.
    5Copyright © 2018Aricent. All rights reserved. The Edge Computing Complexity for an Application Developer Aricent Confidential The application developer has no access to the edge. It is deep inside the service provider network Potentially hundred of application instances running across several edges. Monitoring and monetizing for a developer is a nightmare The telco network is too complex and needs abstraction. NFV model of application deployment may be too complex for application developers Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
  • 6.
    6Copyright © 2018Aricent. All rights reserved. The Developer Experience and a Market Place 5G & Edge Market Place-as-a-Service Onboarding Service (Developer onboarding portal) Developer Services (MEC Developer SDK, Developer Portal) MEC Infrastructure Services Business Model Realization Marketplace-as-a-service enables operators to onboard developers by providing seamless developer experience (DX) 5G/Edge will not be cheap. Software Architecture needs to provide models to rationalize developer’s cost to develop and host applications. Applications can be developed using pay-as-go models like functions-as-a-service / serverless / event based architectures. This means developers pay only when it is used and does not host the application 24*7*365. Edge Data Center Edge Data Center Edge Data Center Edge Data Center Edge Data Center Edge Data Center Edge Data Center Operator & Developer Monetization Application Developers Application Developers (applications, functions) Micro-service Developers (micro-services) 3rd Party App Provider (applications, functions) Distributed Edge Platform #1 Distributed Edge Platform #2 Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
  • 7.
    7Copyright © 2018Aricent. All rights reserved. Distributed Edge and 5G Platform | Edge Compute Ultra-reliable, low-latency high speed communications in 5G is expected to enable mobile digital solutions in healthcare, autonomous cars, advanced robotics use-cases, augmented and virtual reality (AR/VR) applications and accelerates the already rapid growth of the Internet of Things (IoT). Data Center eNB Data Center SGW Data Center PGW App Data Center eNB Data Center SGW PGW App Data Center eNB SGW PGW App L1 L2 L3 App Massively Distributed Platform with Horizontal Scaling and Hierarchical Offloads Source: Shamik Mishra, Vishal Gupta, Raveen Sharma. Decomposable Distributed Architectures for low latency sensitive applications. 2018. (Under publication) Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
  • 8.
    8Copyright © 2018Aricent. All rights reserved. Accelerated and Resilient Distributed Edge Storage 3,6 8,P 2,5 8,9 1,5 6,7 DATA 1 2 3 4 5 6 7 8 9 P Generate Parity 4,7 9,P 1, 2 3,4 • Improvement in storage efficiency, Latency • Employ new generation Clay Code. • Clay Code has • Least possible storage overhead • Least possible repair bandwidth and disk read • Shown 3x repair time reduction and up to 30% and 106% improvement in degraded read and write with CEPH • Acceleration of Erasure Coding Offloading the computation to GPU • Integrate the accelerated erasure code algorithms to CEPH Accelerated Storage to ensure reconstruction and resilience of data over highly distributed edge. Data stored in encoded and distributed blocks and can be reconstructed even if some edge is not accessible for a highly mobile client Source: Dinesh Bhaskaran, Aricent. Accelerated Erasure Coding: The New Frontier of Software-Defined Storage. 2018 Storage Developer Conference Edge: For Caching / CDN / Edge Storage use cases, data can be stored in different geo-locations. Reconstructing data even when certain edges are unavailable is critical to the overall performance. Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
  • 9.
    9Copyright © 2018Aricent. All rights reserved. Model Driven Distributed Applications Hardware Container Cluster Service Mesh for Micro-Services (Scalable, Secure) fn1 fn3fn2 Time fn1 fn2 fn3 Low Latency -> edge Cloud DeviceDistributed App Composer (Model Driven Application Composition) App Model App Configuration fn1 fn2 fn3 AppDistributed Deployment fn1 fn2 fn3 FaaS Event Local Diagnostics Ensconce Management Node Dashboard Cloud Edge Goals for the Approach • Model drive application distribution and function composition • Approach for Stateful functions • Service Mesh for local diagnostics and “finding the needle in the haystack” • Event driven architecture to define computing offload • Auto-Configuration, scale-to-zero • Infrastructure agnostic • Distributed AI / Data Pipelines on container based infrastructure Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
  • 10.
    10Copyright © 2018Aricent. All rights reserved. Aricent Multi-Access Edge Computing Framework Reference Architecture Edge Host Edge Network Function Platform Edge Application Platform Platform & Application Management Edge App Edge App VNFs Cloud IaaS OSS / BSS Application Orchestration Multi-Edge Management 5G and Edge Market Place as a service (Developer Experience DX) Cloud App Cloud App Cloud App Cloud App Cloud App Cloud App ENSCONCE Distributed Edge Platform NetAnticipate (Intent Based Networking) Client App Low Latency Access to Apps Distributed Edge Host Management Orchestration Policies Billing Application Control Platform Management MANO Platform (Network Service Orchestration) Public Cloud Operator Cloud Operator Edge Network Automation and Orchestration Programmable Network Fabric EMLEE (Edge Machine Learning Enhanced Environment) Integrated VNF Frameworks (Cloud-RAN, vEPC) Edge PaaS Edge Application SDK DevOps and Platform Bootstrap Multi-Host Orchestration and Edge Discovery Cloud Native Application Platform Serverless and compute offload Aricent Open Source Projects for Edge Provisioning https://github.com/cablelabs/snaps-boot https://github.com/cablelabs/snaps-openstack https://github.com/cablelabs/snaps-kubernetes Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
  • 11.
    11Copyright © 2018Aricent. All rights reserved. Key Objectives of Aricent Edge Compute Reference Architecture Developer centric architecture for edge applications with edge application SDK Distributed and Event based Computing Architecture to offload computing from devices to edge through Serverless and Sidecar design patterns for edge applications with edge application SDK Multi-Edge Management mechanism enabling devices (and developers) to discover the “right” edge while executing applications Orchestration system to manage the lifecycle of highly available network functions and edge applications with dynamic recovery, application migration & security Automated Bootstrap Complete automated bootstrapping of micro-data centers, platform-as-a-service and management systems with minimal user intervention Accelerated Compute and Storage GPU acceleration for low-latency compute. Distributed edge storage through GPU accelerated erasure codes Drivers for Edge Compute Questions from App Developers The Complexity of Edge Compute Developer Experience Distributed Edge Edge Storage Compute Offload Reference Architecture
  • 12.
    12Copyright © 2018Aricent. All rights reserved. Thank You Headquarters 3979 Freedom Circle Santa Clara, CA 95054 www.aricent.com Copyright © 2018 Aricent. All rights reserved. Shamik Mishra AVP – Technology & Innovation