1 of 31

The “Emerging” Edge

Edge Services, Edge Computing, Flexible Service Orchestration & Dynamic Infrastructure Overlays

Joe Breen

University of Utah

​

Snowmass Computational Frontier Workshop

Aug 10, 2020

1

2 of 31

Where does the science need to happen?

  • On the instrument?
  • At the workstation?
  • At the High Performance Computing Center?
  • In the Cloud?
  • At the “edge” of the network?

​

2

3 of 31

Where does the science need to happen?

  • On the instrument?
  • At the workstation?
  • At the High Performance Computing Center?
  • In the Cloud?
  • At the “edge” of the network?

​

  • All of the above

​

3

4 of 31

The “Edge” is emerging as an area of rapid growth

​

  • Changes are happening in both the industry and science edge
    • These changes intertwine and supplement each other
  • New and rapidly evolving technologies are enabling different approaches
    • WiFi6 - more bandwidth, increased wireless scheduling
    • 5G - more bandwidth, lower latency, network isolation, private cellular infrastructure on CBRS bands
    • Network Function Virtualization - Virtual networks, migration and scaling of virtual network functions
    • Containers - Docker, Singularity, Shifter - ability to package one or more processes in self-contained environments
    • Container orchestration tools such as Kubernetes, Docker Compose -- ability to deploy multiple processes together to create composable services

4

5 of 31

The “Edge” is emerging as an area of rapid growth

  • New workflows and workflow requirements
    • DevOps - developers are increasingly more responsible for the complete software lifecycle, from development to deployment
    • Groups want to create mash-ups of software technologies to create workflows
    • Reproducible and Repeatable science requirements - Scientists want to package their workflows with all of the respective dependencies - move between cloud, HPC, and local resources
  • Scaling workflows
    • LHC experiments
    • Plant Genomics
    • Increased number of sensors in IoT deployment

​

5

6 of 31

The “Edge” is emerging as an area of rapid growth

  • There are two distinct and closely coupled aspects of “Edge”
    • Edge Services
    • Edge Computing

​

6

7 of 31

Edge Services

  • What are Edge Services?
    • User facing services that need to tie tightly to the local compute and storage resources
    • User facing services that need to be as close to the end user as possible in order to maintain specific network and security characteristics
      • Really about maintaining the user experience
    • Services which might benefit from “multi-cloud” capabilities
    • Persistent user managed application services
    • Service methods to increase the resiliency strategy https://www.networkworld.com/article/3280232/living-on-the-edge-5-reasons-why-edge-services-are-critical-to-your-resiliency-strategy.html

​

7

8 of 31

Edge Services examples

  • Persistent services
    • Science Gateways
    • Web portals
    • Reverse proxy and local web caching engines
    • Workflow management systems for specific disciplines
  • Services which need to tie tightly to compute, network and storage
    • Jupyter, Matlab and Mathematica analysis notebooks and portals
    • Data Transfer services
    • Data caching

​

8

9 of 31

Industry Content Delivery Networks

  • Akamai
  • Google Global Cache (content)
  • Google Cloud CDN (service)
    • Hosting web and other content for entities
    • https://cloud.google.com/cdn
  • Netflix Open Connect

​

9

10 of 31

Science Examples of Content Delivery Networks�

10

U.S. ATLAS is using SLATE to deploy XCache for XRootD-based data caching in the US and Europe.

Ilija Vukotic, UChicago

Edgar Fajardo, UCSD

The Open Science Grid operates a world-wide network of XCache-based caches, used by LIGO, Virgo, DUNE and other scientific communities.

11 of 31

Edge Computing

  • What is edge computing?
    • Compute, storage, and network resources that exist at the edge of the network close to the resources
    • Compute, storage, and network resources that exist at the edge of the network close to the user
      • Supports End User Experience requirements of specific network and security characteristics, i.e. low latency, minimum exposure to the internet, distributed availability and resiliency
    • Resources that support specific edge services (usually dedicated)
      • Services sometimes centrally managed by another entity

​

11

12 of 31

Oakridge Leadership Computing Facility

  • Support persistent workflows that leverage local resources
    • Deploying Science Gateways
  • Support user/developer deployed workflow systems
  • Support Continuous Integration / Continuous Development

​

12

https://www.olcf.ornl.gov/2020/03/27/olcf-offers-new-workload-capabilities-with-slate-service/

13 of 31

NERSC

  • Ability to “spin up” services on Rancher (Management system for Kubernetes)
  • Foster “Build, Ship, Run” within scope of NERSC HPC infrastructure
    • Create your own application stack in Docker containers
    • Bundle all dependencies within the containers
    • Mount local NERSC filesystems
  • Support long-term, constantly running services which require persistency and ability to scale
    • Example applications:
      • Frontier Squid Cache
      • CVMFS
      • License servers
      • Science Gateways
      • Jupyter analysis portals

​

13

https://www.nersc.gov/systems/spin/

https://docs.nersc.gov/services/spin/

14 of 31

Edge Services in the Cloud

  • All provide “Edge Services” in the Cloud
    • Support persistent user applications with access to cloud compute, network, and storage
    • Support containers, microservices allowing packaging of dependencies
    • Support large scalability

14

15 of 31

Edge Computing near the end user or experiment

  • Industry example: 5G and autonomous vehicles
    • Central offices and cell towers are converting into micro-datacenters to house Edge compute
    • Edge Compute necessary to support real-time responses and sensor streaming for autonomous vehicles
  • Science Example: Edge computing at the remote experiment
    • Raspberry Pi with SenseHat on space station supporting primary and secondary education
      • https://astro-pi.org/

15

16 of 31

Rapid growth of Edge + Shifting paradigms of software delivery

  • Edge Services and Edge computing is naturally distributed and larger scale
    • Large number of software management points
  • Software delivery is moving towards Continuous Integration / Continuous Deployment
    • In the past, Netscape browser upgrades were local, manual upgrades
    • Today, Google, Mozilla and Apple continually release new versions and push them through central update mechanisms
  • Software deployment is moving more towards DevOps where developers have more responsibility for the full software lifecycle
    • Developers are the experts of the software deployment where ever it may land: cloud, edge, end-user machine -- developers can rapidly see and fix issues

​

16

17 of 31

Rapid growth of Edge + Shifting paradigm of software delivery

  • Software deployment is happening on hardware owned by others
    • At home, you may own a Nest Thermostat, Ring doorbell, Google Home device or Amazon Alexa device but do you handle the software updates?
    • AT&T plans to host software from multiple vendors in their micro data centers while providing isolated 5G networks for the deployments
  • Security requirements are driving the curation of software and deployment of only known validated updates and services
    • Microsoft Azure Sphere now provides hardware, OS and cloud components for deploying IoT --
    • SLATE ci providing a curated catalog of known vetted science software to deploy

​

17

18 of 31

Supporting multiple tenants at the Edge

  • Industry has financial contracts, legal recourse and proprietary rules of use
    • Each entity has terms and conditions by which to abide
    • The owner of the underlying hardware and infrastructure software can charge the entity providing the services -- Telco, Cloud or Edge provider
    • Each entity operates their own layer or portion of the edge.
  • Science has collaboration, federation models and the need for transparency
    • Science still has security requirements and policies to which to adhere
    • Social and security standards need more development for allowing third party deployment of software at edge sites
      • WLCG security working group exploring this topic
      • SCIv2 framework

18

19 of 31

Examples of Federated Operations�Gateways

19

Harvester

U.S. ATLAS and U.S. CMS are both investigating or developing SLATE-based solutions for sending workloads to the Frontera supercomputer at TACC.

SLATE @ TACC

OSG Compute Element

20 of 31

Examples of Federated Operations�Content Delivery Networks

20

U.S. ATLAS is using SLATE to deploy XCache for XRootD-based data caching in the US and Europe.

Ilija Vukotic, UChicago

Edgar Fajardo, UCSD

The Open Science Grid operates a world-wide network of XCache-based caches, used by LIGO, Virgo, DUNE and other scientific communities.

21 of 31

Tying the Edge, Service delivery and support of multiple tenants together

  • Allow deployment of Edge Services and Edge Computing for multiple tenants
    • Ability to deploy in the Science DMZ next to the HPC resources
    • Ability to deploy closer to the user and the instrument
    • Ability to deploy in cloud next to cloud resources
  • Allow deployment of known services in a continuous and secure fashion
  • Allow deployment of federated services
  • Allow federated operations of services
  • Allow deployment of persistent services

21

22 of 31

Federated Operations�Your Services

22

SLATE is designed to allow you to build your own multi-institutional scientific computing collaborations and deploy your services around the world in minutes.

​

For Developers: Design new services, deploy, rapidly iterate.

​

For Sites:

Allow trusted entities deploy applications, but retain control over who and what.

New Services

Award #: 1724821

https://slateci.io

23 of 31

23

Application deployment that is:

  • Rapid
  • Consistent
  • Secure

​

Built with Security baked in

  • Trust relationships and responsibilities
  • Security Policy and Posture Definitions
  • Application Curation Workflow

Software Catalogs

  • Distributed
  • Scalable
  • Manageable

Award #: 1724821

https://slateci.io

24 of 31

Options for developing/testing Edge Services with at scale workflows - National Testbeds

​

  • GENI - distributed compute racks
    • http:s//www.geni.net
    • Utilizes Internet2 OESS orchestration software for WAN orchestration
  • Cloudlab - Cloud testbed and workflows at scale
    • https://cloudlab.us
    • Utilizes Internet2 OESS orchestration software for WAN orchestration
    • Utah/Wisconsin/Clemson and GENI resources
  • Chameleon - Cloud testbed and workflows at scale
    • https://www.chameleoncloud.org
    • Utilizes Internet2 OESS orchestration software for WAN orchestration
    • Computation Institute (Chicago)/iCAIR/TACC/RENCI and GENI resources

24

25 of 31

National Testbeds

  • POWDER - 5G/Wireless testbed
    • https://www.powderwireless.net
    • Develop wireless protocols, test wireless IoT, workflow and edge service development/ deployment in terms of wireless technologies
  • COSMOS - 5G/Wireless testbed
    • https://www.cosmos-lab.org/
    • Develop wireless protocols, test wireless IoT, workflow and edge service development/ deployment in terms of wireless technologies
  • AERPAW - Aerial wireless experiment platform
    • https://aerpaw.org
    • Develop 5G/wireless protocols integrated with drones and other mobile devices in order to improve coverage, signal and location data on fully mobile platforms

​

25

26 of 31

FABRIC

26

27 of 31

FABRIC - tying Testbeds and CI infrastructure together

​

  • Interconnects Testbeds, Computation facilities to support long running, at scale workflow experiments (NOT production domain science)
    • Support research in Cyberinfrastructure and Security
  • Programmable core network infrastructure
  • Programmable core racks
  • Programmable edge racks
  • Guaranteed quality of service
  • Highly accurate timestamping
  • Support of bare metal/VMs/Containers
  • FPGA and SmartNIC support
  • Full visibility at each layer and through the entire network

27

https://fabric-testbed.net

28 of 31

Summary

  • Science happening in multiple locations -- Edge is the current emerging growth area
  • Edge Services/Compute are either close to the resources or close to the user/experiment
  • Software paradigms are shifting to Continuous Integration/Continuous Deployment with developers moving to a full lifecycle DevOps support model.
  • A Federated Operations model supports multi-tenant services on 3rd party edge hardware in the science domains
  • Security policies and models need to continue to develop to support the new operations models.
  • FABRIC will tie national testbeds and national compute together for development of long running workflow scenarios and at scale scenarios

28

29 of 31

thanks

https://slateci.io

29

30 of 31

Explain

  • the paradigm shift of who operates the services vs who owns the hardware. Acknowledge the social challenges.

Give

  • examples of network caches in the network backbone, and other services

Workflow

  • management / Facility-apis (for all compute and storage not just for ‘analysis’ or edge-services)

K8s & Helm - Declarative infrastructure��Expanse

  • as an example of joint Slurm & K8S facility including cloud scale out

�Plethora

  • of other tools/approaches - Dask; Ray; Lambda step-functions ..

​

30

31 of 31

  • Where does the science need to happen?
  • Edge services
    • What are edge services
    • Example: OCLF Slate
    • Example: NERSC SPIN
    • Example: Edge services in cloud -- whole AWS, Google and Azure premise
  • Edge computing
    • What is edge computing
      • Example: CDNs
      • Example: 5G and automotive
      • Example: Physics with sensors -- staging compute/storage or latency sensitive
  • Deliver software on other people’s hardware?
    • Every group might have edge services but do they also have to deploy edge compute AND keep the software up to date?
    • Need a change in mindset (or do we just need to expand?) and need to address security
    • Example: Netscape manual upgrades vs Google, Mozilla, Apple constant delivery and updates -- continuous deployment / continuous integration
    • Example: Nest thermostat or Ring doorbell or Google Home or Alexa
    • Example: Microsoft and Ford?
    • Example Microsoft IoT - curated /hardened software
    • Example of software on Mars Rover or space station with Raspberry Pi/SenseHat - curated software
  • SLATE - combining these thoughts
    • TACC example - Edge services
    • Edge computing -- target small schools or in the field where IoT is
    • Software delivery from curated platform
    • Working with WLCG working group on security policies
  • Designing the full ecosystem with full visibility What are the next advances?
    • GENI, Chameleon, Cloudlab Testbeds - Create workflows, orchestrate movement of compute
      • Slices between compute blocks but limited programmability and visibility in middle
    • 5G - POWDER - sliced network, isolated services
    • FABRIC - programmability and visibility at all layers - edge services, edge computing, cloud computing, and fully programmable/visible middle
      • Developing with technologies that came from GENI, Chameleon and Cloudlab

31