Dell PowerScale: Considerations and Best Practices for Large Clusters

• 1

  Overview
• 2

  Audience
• 3

  Revisions
• 4

  We value your feedback

• 1

  Cluster definitions

• 1

  Considerations for planning and designing a large cluster
• 2

  Cluster architecture
• 3

  Single large cluster
• 4

  Storage pod

• 1

  Workload analysis
• 2

  How does the application work?
• 3

  How do users interact with the application?
• 4

  What does the network topology look like?
• 5

  What is the application’s performance profile?
• 6

  What are the application’s I/O requirements?
• 7

  What is the disk latency?
• 8

  How much CPU utilization?
• 9

  When to analyze workloads?

• 1

  Large cluster health check
• 2

  Core sanity check
• 3

  Disk usage
• 4

  CPU and memory usage
• 5

  Log files
• 6

  Core services

• 1

  Node hardware recommendations
• 2

  Physical and environmental design
• 3

  Extra-large cluster design example
• 4

  Power and cooling
• 5

  Cabling and networking

• 1

  Cluster administration and management considerations
• 2

  Cluster upgrades
• 3

  OneFS hardening
• 4

  Cluster informational resources and Dell Support
• 5

  Cluster monitoring and auditing
• 6

  InsightIQ
• 7

  Secure Remote Services
• 8

  Engaging Technical Support

• 1

  Cluster capacity considerations
• 2

  Maintaining appropriate protection levels
• 3

  Monitoring cluster capacity
• 4

  Configuring storage quotas
• 5

  Scenario 1: quota management
• 6

  Scenario 2: HPC compute farm constraint

• 1

  Cluster composition and group state
• 2

  Minimum access level for capabilities per node state
• 3

  Minimum access level for capabilities per drive state
• 4

  Understanding and analyzing group membership
• 5

  Identifying accelerator nodes
• 6

  SmartFailed and down nodes
• 7

  SmartFailed and down drives
• 8

  Read-only nodes
• 9

  Dead nodes
• 10

  Dead drives
• 11

  SmartFailed and stalled drives
• 12

  Reading group sequence numbers
• 13

  Group changes
• 14

  Interpreting group changes
• 15

  Constructing an event timeline
• 16

  Extra-large cluster example

• 1

  Layout, protection, and failure domains
• 2

  Tiers
• 3

  Failure domains and neighborhoods
• 4

  File layout and directory structure

• 1

  Data layout and tiering recommendations

• 1

  Multi-tenant recommendations

• 1

  Job engine recommendations
• 2

  Job engine monitoring

• 1

  Data availability, protection, and disaster recovery considerations
• 2

  Replication recommendations

• 1

  Best practices checklist

• 1

  Summary