agile hpc with vmware cloud foundation & ubercloud€¦ · grid engine, kubernetes) ... app os...
TRANSCRIPT
1 Confidential │ ©2020 VMware, Inc.
Agile HPC with VMware Cloud Foundation & UberCloud
Thomas Francis, UberCloud
Mohan Potheri, VMware
Confidential │ ©2020 VMware, Inc.
Agenda
2
• Traditional HPC architecture
• Path to agile HPC
• vSphere announcements for Kubernetes
• Containers and HPC with UberCloud
• Example deployment
Confidential │ ©2020 VMware, Inc.
Traditional HPC Architecture
3
• Login / Head Node
• Scheduler
• Storage Gateway
• NFS Storage
• Compute Nodes
• Accelerator Nodes
• Parallel Filesystem (Lustre)
• High Speed Interconnects
Confidential │ ©2020 VMware, Inc. 4
Variety of Devices (CPU, RAM, Storage, Networking/RDMA, GPU, FPGA)
Workload Management Options (Moab, PBS, Grid Engine, Kubernetes)
Dormant infrastructure that is only refreshed every 3 years
Slower innovation
Complex resource and application management
Higher Costs
Siloed operations
80% of budget spend on keeping the lights on
Complex Infrastructure Business Impact
Enterprise HPC Today
4
Confidential │ ©2020 VMware, Inc.
Virtual Machine Benefits
5
hardware
hypervisor
VM OS App
Virtual Machines offer: • Multi-cluster aggregation
(heterogeneity)
• Multi-tenant data security
• Fault isolation
• Reproducibility
• Fault resiliency
• Dynamic load balancing
• Performance
hardware
Confidential │ ©2020 VMware, Inc. 6
The Road to a Modern Data Center Architecture for HPC
6
HW
Legacy Apps
Cloud Native Apps
From traditional Data Center to unified, scalable Data Center operations…
Combining existing tools with new data center technologies • VMware VCF virtualizes Data Center resources • Kubernetes is the orchestrator • Docker Containers are the new kernel • UberCloud Containers bridge to HPC • Enabling HPC and AI to converge • Preserving existing CAPEX investment • Addressing rapidly evolving business requirements • Accelerating the HPC application lifecycle
Confidential │ ©2020 VMware, Inc. 7
VMware vSphere 7 & VCF 4: Unifying Kubernetes and VMs vSAN
IT Operator Developer
Kubernetes is embedded into the control plane of vSphere
vCen
ter S
erve
r
AI/ML Business Critical
DB & Analytics Time-critical
Namespaces
VMware Cloud Foundation Services
Tanzu Runtime Services
Hybrid Infrastructure Services VM Service Storage Service Registry Service
vSphere Pod Service Network Service Hybrid Mobility Service
Tanzu Kubernetes Grid Service
vSphere vSAN NSX vRealize
Confidential │ ©2020 VMware, Inc.
VMware + UberCloud Provide:
8
• Hybrid and multi-clouds
• HPC, Big Data, AI, Enterprise apps and microservices
• Commercial, OpenSource, in-house codes
• CAE, Life Sciences, EDA, Finance, Oil & Gas, …
Choice Agility Security
• Unified operations • Rapid delivery and
response • Cost efficiency
• HW flexibility
• Package once, move to any cloud
• On-demand Scaling of resources
• High Availability & Auto-Recovery
• Safer applications • Governance • Chain of custody
• Threat mitigations
• Multitenancy
• HW+Network+SW isolation
• Additional Login/Password for app container protection
• iLO with Trusted Platform Module offer enhanced security
Confidential │ ©2020 VMware, Inc.
Containers: The New Application Deployment Standard
Application
Operating System
Run Time
Middleware
Infrastructure Integration
• Portable across hybrid infrastructure
• Continuous updates
• Can bridge HPC & Enterprise IT
Cloud Infrastructure
Confidential │ ©2020 VMware, Inc.
Automated Container Factory: CI/CD to build containers
• Templatized HPC containers
• Automated Build
• Integrated Testing
• Controlled Versioning
10
Confidential │ ©2020 VMware, Inc.
Deployment
11
Capabilities On Premise Off Premise* Public Cloud Interactive Access (allowing full engineering workflow)
✔ ✔ ✔
Batch Access ✔ ✔ ✔
Data Isolation (by user/group)
✔ (per cluster)
✔ (per cluster)
✔ (per cluster)
Access Security ✔ (inside network)
✔ (subnet & VPN)
✔ (VPN)
Integrated Security ✔ ✔ ✔ (Automated Threat Scanning, Controlled Code Deployment, Encrypted Communications, Role-based access
control, Policy Management and Enforcement, Image Signing)
RDMA Support ✔ ✔ ✔
GPU Support ✔ ✔ ✔
Cluster Resizing ✔ ✔ ✔
Multi-User Access ✔ ✔ ✔
On-Demand Access ✔ ✔ ✔
Single Image on platforms ✔ ✔ ✔
Certification ✔ ✔ ✔ (Certified plug-ins and Infrastructure, Certified ISV apps as containers, Certified professionals)
Automation ✔ ✔ ✔ (Built-in app reliability/High Availability, Policy-based automation, Auto health check, Auto Load Balance/Deploy/
Recovery, Containerized Application Lifecycle Management, Automated Deployment/DevOps)
Orchestration ✔ (UGE, K8s, others)
✔ (UGE, K8s, others)
✔ (UGE, K8s, others)
Add, remove compute nodes (containers), or terminate service.
Execute Docker Run to start Login Container and Compute Containers
Configure multi-node container environment
Load UberCloud Docker Container Image on all VMs
Pull UberCloud Docker Container Image from Registry to Cluster Storage
Install Docker EE (Runtime)
Configure virtual network, devices, and storage
Create new cluster in vSphere and add hosts login and compute resources
Deploy VCSA and Configure vCenter (management server(s))
Deploy ESXi on servers (management and compute)
Confidential │ ©2020 VMware, Inc. 12
Video Demo
12
Confidential │ ©2020 VMware, Inc.
HPC Case Studies
13
50 HPC Case Studies with UberCloud’s HPC Containers since 2015
• Selection of 21 Containerized Engineering Case Studies: https://info.theubercloud.com/ubercloud-container-compendium
Using HPC Software Containers based on Docker
Confidential │ ©2020 VMware, Inc.
Thank You
14