StarlingX - Driving Compute to the Edge with OpenStack
1. StarlingX
Driving Compute to the Edge With OpenStack
Pilot Project Supported by the OpenStack Foundation
Greg Waines, WIND RIVER SYSTEMS
Architect
2. What Is Driving Edge Computing?
Devices /
Things
MANUFACTURING
TRANSPORTATION
ENERGY
VIDEO
HEALTHCARE
RETAIL
Core
Network
Cloud
Data Center
Network Hub or
regional Data Center
Edge
Compute Node
Varies <1 ms <5 ms <10-40 ms < 60 ms ~100 ms
DRONES
PHONES
SMART
CITIESPCs
Drivers for edge
Latency
Bandwidth
Security
Connectivity
Latency
âWhereâ Matters for New Genres of Applications
3. Internet
of Things
Augmented
Reality
Accelerated
Browsing
Virtual Reality &
Gaming
Vehicle
Assistance
Medical
Applications
NFV
RAN
SCALE
LARGE AND SMALL
CENTRAL MGMT
ACROSS MULTIPLE EDGE CLOUDS
Edge Computing Challenges
ZERO TOUCH
PROVISIONING
SINGLE PANE
OF GLASS
EDGE CLOUD
AVAILABILITY
& AUTONOMY
SECURITY
IN NON-PHYSICALLY
SECURE ENVIRONMENTS
4. An OpenStack Pilot Project
StarlingX is being piloted under the administration and governance of the
OpenStack Foundation with an Apache 2.0 license
⢠Formed with seed code from the Wind River Titanium Cloud portfolio
⢠Open sourced with the imperatives of
- High Availability,
- High Performance & Scalability, and
- Edge-Enabled,
... in a fully-integrated deployment-ready solution,
⢠First community release of StarlingX is planned October 24th, 2018
⢠OpenStack invites new project members and contributors as we build a
mission critical infrastructure stack for Edge computing
⢠Web Site: StarlingX.io
5. StarlingX - Use Cases
⢠Drones, cars, trucks
⢠Robotics, vPLC
⢠Cloud/virtual RAN
⢠Smart city/buildings
(metering & monitoring)
⢠Mobile HD video
⢠Healthcare imaging &
diagnostics
⢠Caching & surveillance
Core
Networks
IoT
Infrastructure
Network
Appliances
Radio
Access
Network
Content
Delivery
Networks
Virtual
CPE
Multi-access
Edge
Computing
Telco Infrastructure
Energy ManufacturingHealth Care
6. StarlingX â Software Architecture
StarlingX provides a deployment-ready,
scalable, highly reliable Edge
infrastructure software platform
Services from the StarlingX
virtualization platform focus on
⢠Easy deployment,
⢠Low touch manageability,
⢠Rapid response to events
⢠Fast recovery
A fully featured Cloud for the distributed Edge
StarlingX.io
8. StarlingX â Standard Configuration
External Networks
Layer 2 Switch
StarlingX
Storage
NodeStorage
Node
Storage
NodeStorage
Node
Control
Node
Control
Node
Compute Node
VM VM VM
Accelerated
vSwitch
Compute Platform
Compute
NodeCompute
NodeComp
ute
NodeCompute
NodeCompute
NodeCompute
NodeCompute
NodeCompute
Node
Compute
NodeCompute
NodeCompute
NodeCompute
NodeCompute
NodeCompute
NodeCompute
NodeCompute
Node
Compute
NodeCompute
NodeCompute
NodeCompute
NodeCompute
NodeCompute
NodeCompute
NodeCompute
Node
Layer 3 Routers
Optional
2-9 Storage Nodes
2-100 Compute Nodes
1:1 HA Control Cluster
OpenStack Cluster deployed on
dedicated physical servers.
â StarlingX build è Installation ISO
â Initial server, controller-0,
becomes installer for rest of system.
§ 2x Node
HA Controller Cluster
§ 2-100x Node
Compute Cluster
§ OPTIONAL:
2-9x Node
CEPH Storage Cluster
9. StarlingX â Range of Scalable Deployments
Highly-Available
Edge Solution
Two servers
Compute
VM
Control
VM VM
Storage
1:1 protected pair
of servers
Compute
VM
Control
VM VM
Storage
Compute
VM VM VM
Compute
VM VM VM
Top of Rack
Control
Storage
Control
Storage
Frame-Level
Solution
4 â100 servers
Compute
VM VM VM
Compute
VM VM VM
Top of Rack
Control
Storage
Control
Storage
Compute
VM VM VM
Compute
VM VM VM
Top of Rack
Control
Storage
Control
Storage
Compute
VM VM VM
Compute
VM VM VM
Top of Rack
Control
Storage
Control
Storage
Large-Scale Data Center Solution
OR
Distributed Edge Computing Solution
Hundreds of servers
Compute
VM VM VM
Compute
VM VM VM
Top of Rack
Compute
VM VM VM
Compute
VM VM VM
Top of Rack
Compute
VM VM VM
Compute
VM VM VM
Top of Rack
Control
Control
Compute
VM VM VM
Compute
VM VM VM
Top of Rack
Compute
VM VM VM
Compute
VM VM VM
Top of Rack
Compute
VM VM VM
Top of Rack
Compute
VM VM VM
Control
Control
Multi-region cloud
Compute
VM
Control
VM VM
Storage
Minimum-Footprint
Edge Solution
Single server
Note:
Supported on Low Cost / Low
Power HW Form Factor.
E.g.
Xeon D-1500 (8 -12 core) Boards
( e.g. Supermicro )
10. Fully Managed Cloud Infrastructure
⢠Simple Installation,
⢠System and Node Configuration ,
⢠âPlug & Playâ mode, and
⢠âBulkâ mode.
⢠Service/Software Configuration,
⢠Service/Software and Node Maintenance
⢠Fault Monitoring, Alarming and Recovery,
⢠Administrative Commands,
⢠Statistics Collection,
⢠External Centralized Log Management.
⢠CLI, GUI, REST API & SNMPv2c Management Interfaces,
⢠Software Patching and Upgrade Management & Orchestration,
⢠Backup and Restore Support.
12. Feature Rich Virtualized Networking
High-Level Networking Options:
§ Virtualized Switching & Routing,
( OVS-DPDK accelerated )
§ PCI-Passthrough and SRIOV,
§ SDN Networking,
Key Data Path Features:
§ Distributed Virtual Routing (DVR),
§ Security Groups,
§ Floating IP and SNAT,
§ VLAN-tagging in Guest,
§ Neutron Service Function Chaining,
§ VM Access to Data Path Acceleration HW
High-Availability:
§ LAG for Link Redundancy / Protection,
§ QoS features for VM Traffic,
â BW Limiting, DSCP marking,
§ Live VM Migration for all Networking
modes,
Network Diagnostics:
§ LLDP Peer Discovery and Reporting,
§ TcpDump-based Packet Tracing tools for
Guests.
13. Enhanced Platform Awareness
Hardware Server
...
cores
memory ...
cores
memoryQPI
Virtual
Machine
Virtual
Machine
Virtual
Machine
VM Memory Backed by 4K,
2M or 1G Memory Pages
NUMA Affinity optimized relative
to vSwitch and/or VMâs PCI-
Passthru/SRIOV Interfaces
No-Over-Commit / Dedicated CPU Model
for Guaranteed VM Performance Assign Scheduling Policy
to vCPUs
CPU Model Specification
for leveraging specific CPU Archs
GPU & Generic PCI Device Passthru
for VM
Isolate or Require
Hyper-threading Policies
Virtual NUMA Node Support
in VMs
Server Group
âCompute-Node Affinity / Non-Affinity
Ensure Compute Node has connectivity to
VMâs attached Tenant Networks
RDT Cache Allocation Technology
to provide VM exclusive access to L3
Cache
14. Container Support
Ironic
Instance
(baremetal)
Ironic
Instance
(baremetal)
⢠OpenStack Magnum
⢠Container Orchestration Engine
management
⢠Containers in VMs,
⢠Via OpenStack Nova
⢠Containers on Bare Metal Servers,
⢠Via OpenStack Ironic
⢠Kubernetes, Swarm and Mesos
⢠Container Runtime : docker
⢠Container Networking : flannel, calico
⢠Container Storage : cinder
OpenStack
Compute Nodes
Container
Baremetal Nodes
Container Runtime
COE
Container
Container Runtime
Ironic
Hypervisor
Nova
Instance
(VM)
Nova
Instance
(VM)
Nova
Instance
(VM)
COE
Container
Container
Runtime
Container
Container
Runtime
HEAT
NEUTRON NOVA
bmlibvirt
Magnum
15. StarlingX Software Management
⢠Software Patching
⢠corrective content and/or minor new functionality,
⢠âALLâ layers of StarlingX Software,
⢠From kernel up to OpenStack Services,
⢠Rolling Update strategy across nodes,
⢠In-service and Reboot-required patches,
⢠Orchestration tool automates patching across âALLâ nodes.
⢠Software Upgrades
⢠From one major release to another major release,
⢠âALLâ layers of StarlingX Software
⢠From kernel up to OpenStack Services,
⢠Rolling Update strategy across nodes,
⢠Transparently handles DB schema chgs and DB data conversion,
⢠Manages API compatibility between nodes,
⢠Live migrates hosted applications,
⢠Orchestration tool automates upgrades across âALLâ nodes.
Cloud Running on Rel N
From Side
Hitless
Migration
Cloud Running on Rel N+1
To Side
Step 3 Upload the patch file to the patch storage area.
Click the Upload Patch button to display an upload window from which you can browse your
workstation's file system to select the patch file. Press the Upload Patch button once the selection
is done.
The patch file is available now in the patch storage area, but it has yet to be applied to the cluster.
This is reflected in the Patches page as illustrated below.
Step 4 Apply the patch.
Click the Apply Patch button associated with the patch. Alternatively, select the patch first using
the selection boxes on the left, and then click the Apply Patches button at the top. You can use
this selection process to apply all patches, or a selected subset, in a single operation.
The Patches page is updated to report the patch to be in the Partial-Apply state.
174 Draft - 27 Jan 2015
16. Distributed Edge Computing
Central Cloud
Region
Edge Cloud
Region
Edge Cloud
Region
Edge Cloud
Region
Layer 3 IP Network
§ Based on OpenStack Regions,
§ Central Cloud Region:
â Hosting Shared Services
â System-wide Infrastructure Orchestration functions:
§ Deployment and Management of Edge Clouds,
§ Configuration portal for shared configuration across all Edge Clouds,
§ Fault aggregation across all Edge Clouds,
§ Patching orchestration across all Edge Clouds.
§ Remote Edge Cloud Regions:
â Geographically dispersed,
â Scalable from 1 to 100s of Servers,
â Connected via L3 IP Network,
â Running reduced Control Plane.
§ Inter-Region Communications strictly REST APIs / L3.
17. In-Flight Evolution of StarlingX
Evolving to a Containerized OpenStack Solution
§ Leveraging OpenStack Helm Project,
to deploy the OpenStack Cloud in Containers on a Kubernetes Cluster.
Value Add:
èFor OpenStack Solution:
Ă Open framework for integrating & installing OpenStack Services,
Ă Kubernetes & Helm == OpenStack Installation & Configuration Service.
èFor Container-ONLY Solutions:
Ă Introduces a **NEW** HA Integrated Managed Kubernetes Solution
Continues to leverage StarlingX Infrastructure Services to:
§ install & manage bare metal servers,
§ install & manage Kubernetes deployment, and
§ orchestrate install of Containerized OpenStack with OpenStack Helm.
Titanium Services
HA,Config,Mtce,EPA,
Security,Patching
Titanium
Services
App
Titanium
Services
App
Titanium
Services
App App
18. Summary End-to-End Integrated OpenStack Solution
Deployment Ready
Distributed Edge Computing Solution
Scales both small and large
Highly Available
HA Controller Services
HA Management of Hosted VMs
Live Migration
Hitless Software Patching and Upgrades
Enhanced Network Performance
Enhanced VM Performance
Scalable from 1-server to 500+ Distributed Cloud
Quick & Easy Installer
Intuitive Configuration & Operations
Fully
Integrated
Solution
Usable &
Simple
Available
& Reliable
High
Performance
& Scalability
Built for
the EDGE
Flexible
Hosting
Virtual Machines
Containers
Bare Metal Nodes
19. Invitation to Join the Community
⢠We cordially invite you to join the StarlingX community
⢠starlingx.io provides links to code
⢠starlingx.io provides links to documentation (soon to be at docs.starlingx.io)
⢠Build Guides, Install Guides, Contributor Guides, ...
⢠Sign up to the mailing list at lists.starlingx.io
⢠Attend community meetings at wiki.openstack.org/wiki/StarlingX#Meetings