Stratus Protected Data Cloud

Stratus Protected Data Cloud

MSI is building a local research compute cloud environment called Stratus (https://stratus.msi.umn.edu), which is designed to store and analyze protected data, such as dbGaP. Stratus is a subscription-based infrastructure as a service that enables users to operate within their own self-service virtual machines on the spectrum between a low-level bare-metal HPC and a fully virtualized, shared-tenancy mode. Stratus is powered by the Newton version of the OpenStack cloud platform, and is backed by Ceph storage

The initial interation of Stratus will feature functionalities not available on other MSI systems:

  • The self-service portal allows users to create virtual machines and remote desktops on demand.
  • Long-running jobs (i.e., > 30 days) are enabled by leveraging live-migration during maintenance windows.
  • Virtual machines support Docker containers for self-service software management (e.g., to deploy local versions of Galaxy or JupyterHub).
  • A secure storage platform provides the necessary isolation and protections for users to process protected data.

NOTE: This system is part of the MSI Beta program and details are subject to change.

 

Hardware

The Stratus compute hardware currently consists of:

  • Twenty HPE Proliant XL230a compute nodes, each with two Intel E5-2680v4 (14-core) CPUs, 256 GB of RAM, and 10 GigE networking to the outside world.
  • An array of eight HPE Apollo 4200 storage servers, connected to compute nodes via redundant 40 GigE switches.
  • Eight additional servers tasked with monitoring storage operations and orchestrating virtual machine lifecycles.
Storage

The initial storage deployment on Stratus has:

  • 200 TB of Ceph block device storage for ephemeral VMs and data volumes
  • 400 TB of S3-compatible Ceph object storage for short-term protected data caching
  • Separate S3 object storage to store persistent protected data copies*,**

* Option available for purchase per-TB/Year

** Persistently stored data must have an original copy stored at another location

Subscription

Access to Stratus is granted by subscription fee paid on a quarterly or annual basis. Prices are subject to change with the University's fiscal year, and subscriptions automatically renew.


A breakdown of the FY19 costs for Stratus services is as follows:

SERVICE NAME UNIT COST/year COST/qtr
Stratus base subscription (*) Pkg $663.40 $165.85
Additional Virtual CPU Cores (incl. +2GB RAM/vCPU) vCPU $20.88 $5.22
Additional Block storage TB $164.69 $41.18
Persistent Secure Object storage TB $76.22 $19.06

 


(*) The base subscription gives users:

  • 16 vCPUs and 32GB RAM for use on one or more virtual machines
  • 2TB block storage for virtual machines and attached volumes
  • access to a virtual machine image repository containing MSI-approved CentOS and Ubuntu images
  • root access to install software on personal virtual machines
  • For Controlled-access Data Users:
    • Access to the S3 protected data cache for short-term storage of controlled-access data (maximum shared capacity: 400TB)
    • Sane default configurations for virtual machines, network, and storage that satisfy controls required by most data use agreements

ADDITIONAL NOTES:

  • Stratus services and subscriptions are currently restricted to University of Minnesota users. 
  • Stratus compute nodes have a maximum of 256GB RAM per node. Each node has 28 physical cores and 2x hyperthreading enabled. These limits dictate the resource limits for any single virtual machine. 
  • RAM is allocated to projects at a ratio of 2GB RAM per vCPU, and is not an oversubscribed physical resource. Increase the available RAM in a project by scaling the number of vCPUs.
  • Users can control how RAM and vCPU resources are independently assigned to virtual machines, but RAM and vCPUs are only sold as a single unit. Thus, allocating more than 2GB RAM per vCPU to a single virtual machine either implies less RAM per vCPU on another machine or unused vCPUs in the project. 
  • vCPUs oversubscribe the physical hardware 4:1 (i.e., 2x oversubscription * 2x hyperthreads per core). Some applications benefit from hyperthreading, but many do not; more than 28 vCPUs per virtual machine is guaranteed to leverage hyperthreading. 

Contact the MSI helpdesk to enroll today:  Email help@msi.umn.edu

 

 

 

Documentation

Documentation and Frequently Asked Questions about Stratus can be found here

Stratus FAQ

Publications

Publications about Stratus can be found below

  • Bollig, E. F., and Wilgenbusch, J. C., From Bare Metal to Virtual: Lessons Learned when a Supercomputing Institute Deploys its First Cloud. In Proceedings of the Practice and Experience on Advanced Research Computing (PEARC 18), 2018. ACM, New York, NY, USA, Article 13, 8 pages. DOI: 10.1145/3219104.3219164
  • Bollig, E.F., Allan, G. T., Lynch, B. J., Huerta, Y. A., Mix, M., Munsell, E. A., Benson, R. M., and Swartz, B., Leveraging OpenStack and Ceph for a Controlled-Access Data Cloud. In Proceedings of the Practice and Experience on Advanced Research Computing (PEARC 18), 2018. ACM, New York, NY, USA, Article 18, 7 pages. DOI: 10.1145/3219104.3219165
  • Bollig, E. F., Allan, G. T., Lynch, B. J., Huerta, Y., Mix, M., Swartz, B., Munsell, E., Leibfried, J., Hospodarsky, N., Managing dbGaP Data with Stratus, a Research Cloud for Protected Data. In Proceedings of the Practice and Experience in Advanced Research Computing 2017 on Sustainability, Success and Impact (PEARC17). ACM, New York, NY, USA, Article 65, 4 pages. DOI: 10.1145/3093338.3104185