Scientific Data Analysis on the Cloud
We recongize the fact our biggest stakeholder is our science communities. This technical session focues on using the Cloud to conduct scientific data analysis. Some of the focuse areas include
1. Technologies and solutions resuse
2. Reference architectures
3. Integration with multi-cloud environment
4. Leverage social networks and contribute back
NASA Earth Exchange (NEX): Community Engagement in the Cloud through OpenNEX
Petr Votava, NASA Ames/CSUMB
Overview of NEX
Discussion on Expanding to the Cloud
The Setup:
-NASA-Amazon Space Act Agreement: 1-year experiment through Nov 2014
Testing and Feedback
-Series of contests around data and services
-NASA Prizes and Challenges Program
--Start w/SpaceApp Challenge
-Improvement
Final Product
-Learned from past lessons
-OpenNEX Virtual Workshop and Challenge 2014
-https://nex.nasa.gov/OpenNEX
The OpenNex Challenge
-Run in collaboration with Innocentive
-Part 1: Ideation (Idea Generation)
-Part 2: Implementation on AWS
What's next: extending SAA and beyond
-Looking to engage other providers
-Work w/NASA on other ways to produce services beyond SAA
What's next: Elastic capacity
-Testing of expanding HPC architecture into the cloud
What's next: Visualization and Analytics
What's next: Easier access to HPC
Some Observations
-Technical Challenges
-Organizational Challenges
-Cultural Challenges
Discussion of a vision for an "Ideal (Open)NEX Platform"
-----
The Virtual Machine Scaler: Infrastructure Management Support for Scientific Modeling on IaaS Clouds
Wes Lloy, Colorado State University
CSIP: Cloud Services Innovation Platform
-Provide scientific modeling as a service (MaaS)
-Facilitate science research and delivery
Supporting Science Discovery and Delivery
-UDSA-AG Systems Model Research
-USDA-NRCS: AG Systems Production Models
CSIP Model Services Diagram
Scientific Modeling Cloud Challenges
-Model Services Deployment: deploy each component of an application to VMs and run in isolation. Not cost effective. Can take advantage of combinations and consolidate VMs.
-Elasticity: Scale computational resources for each tier of the application
-Green computing can lead to resource contention
-Overprovisioning: too many VMs on same host
-Virtualization Overhead
Discussion of Amazon Spot Instances
The Virtual Machine (VM) - Scaler
-Organization Diagram showed
-Web services application
-Discussion of Supporting features
VM Pools
-Supports work with many same-type VMs
-Addresses Launch Latency
Resource Utilization Data Collection
-Resource utilization sensors
--Sensor on each VM/PM
--Transmits data to VM-Scaler at configurable intervals
Resource Utilization Checkpointing
-Captures resource utilization at a time
Scaling Tasks
-Scaling service request
-Prelaunch VMs
-etc
Hot Spot Detection
-Resource utilization thresholds
-Performance model approach
Least-Busy VM Placement
Least-Busy Job Placement
-----
Eucalyptus 4.0 Upgrade
Brian Thomas
Gives IT and DevOps teams the power they need to easily deploy and manage large-scale clouds.
Discussion of Key Advancements in 4.0
Edge Networking
Discussion of AWS Compatibility
-----
Cloud Computing Cluster onward
Phil Yang
Cloud computing cluster was formed 3 years ago
Discussion of cloud computing experiences among members in attendance
-----
Other Notes:
2014 AGU Meeting
-Session 3041 and Session 1832
NASA ESDWG (Earth Science Data Working Group)
Comments
Session Presenter - Petr Votava - NASA Earth Exchange (NEX)