Title: Transition in Campus CyberInfrastructure: Community Clusters, Storage and CoLoc
1Transition in Campus CyberInfrastructure
Community Clusters, Storage and Co-Loc
Presented by Dwight McKay, Director of Systems
Engineering ITaP Rosen Center for Advanced
Computing Purdue University
2Outline
- Introduction
- Community Clusters
- Summarize model and issues
3Introduction
Community Clusters
Summary
4Introduction
Community Clusters
Summary
5Rosen Center for Advanced Computing
Introduction
Community Clusters
Summary
6Rosen Center for Advanced Computing
Introduction
Community Clusters
Summary
7Transition in Research Computing Support
Introduction
Community Clusters
Summary
8Transition in Research Computing Support
- Change in Direction
- From central purchase to researcher / project
purchase - From central shared facility to resource or
service provider - From service desk to partner
Introduction
Community Clusters
Summary
9Transition Implications
- Paying Customers
- Higher Expectations
- Service Support
- Formal Agreements
- Cultural Change
Introduction
Community Clusters
Summary
10Rosen Center for Advanced Computing
Research
User Support
Infrastructure
Introduction
Community Clusters
Summary
11Rosen Center for Advanced Computing
Project
Project
Introduction
Community Clusters
Summary
12Rosen Center for Advanced Computing
Project
Introduction
Community Clusters
Summary
13Transition Implications
- New Business Models Needed
- HW/SW, infrastructure, support?
- Unpredictable Demand Funding
- Planning for power/space/cooling?
- Non-paying Users
Introduction
Community Clusters
Summary
14Community Clusters
- Condominium Model
- Purchase Computation by the node
- Nodes come with bundle of services
- Purchase Storage by the TeraByte
- Storage comes with bundle of services
- Custom Arrangements for Specific Projects
Introduction
Community Clusters
Summary
15Community Clusters
- Disk Storage
- Archival Storage
- On-Call Support
- Disaster Recovery
- Security
- HW Installation
- HW Maintenance
- Facilities Support
- Network Connection
- Infiniband Connnection
- OS installation and management
Introduction
Community Clusters
Summary
16Community Clusters
- Disk Storage
- Archival Storage
- On-Call Support
- Disaster Recovery
- Security
- HW Installation
- HW Maintenance
- Facilities Support
- Network Connection
- Infiniband Connnection
- OS installation and management
Node Bundle
Introduction
Community Clusters
Summary
17Community Clusters
- Tiered Cycle Allocation
- Owners guaranteed specific share
- Owners given first pick of idle cycles
- Owners agree to harvesting remaining idle cycles
- Use the whole buffalo. -- Brad Bird
Introduction
Community Clusters
Summary
18Community Clusters
Owner
Introduction
Community Clusters
Summary
19Community Clusters
Owner
Pre-empt
Introduction
Community Clusters
Summary
20Community Clusters
Owner
Cycle Harvesting
Pre-empt
Introduction
Community Clusters
Summary
21Community Clusters
- Challenges
- Business model that spans system generations
recovers shared infrastructure costs - Cluster Heterogeneity
- Multiple communities sharing one cluster
(TeraGrid, OSG, NWICG) - Other architectures special needs?
Introduction
Community Clusters
Summary
22Community Clusters
- Storage
- Three Primary Tiers
- Fast for scratch files
- Commodity for home directories
- Archival for longer term storage
- Custom Storage for Specific Projects
Introduction
Community Clusters
Summary
23Community Clusters
- Storage Challenges
- Business model that spans lifetime of data /
researcher? - Media Purchase vs. Space Rental?
- Data Retention Policy
Introduction
Community Clusters
Summary
24Community Clusters
- Connectivity Challenges
- Multiple classes of network are needed
- Direct Routes
- Data Center to Key Research WANs
- Data Center to Key Research Labs
Introduction
Community Clusters
Summary
25Introduction
Community Clusters
Summary
26Community Clusters
- Four Fold Network
- Commodity
- Secure
- Research
- High Performance / Large Data
- Network Research
- Low Latency
Introduction
Community Clusters
Summary
27Summary
- Transition to resource provider / partner /
service - Architecture
- Computation -gt Community Clusters
- Storage -gt 3 Tiers
- Connectivity -gt Direct Data Center connections to
Research WAN and Lab Access - Challenges
- Customer expectations and implications
- Serving up storage and other architectures by
the slice - Recovering ALL the costs, especially support
Introduction
Community Clusters
Summary