I85PREP
⌘ K
Dell PowerScale Deploy study materials
Dell PowerScale Deploy study resources

Dell PowerScale Deploy: Cluster Planning and Network Design

Dell PowerScale Deploy · Work through your exam, one topic at a time.

All guide sections

Cluster Planning and Network Design

Domain overview

Cluster Planning and Network Design tests whether you can size a cluster and design front-end, back-end, and management networks that OneFS will actually support. Questions often hinge on protection overhead, IP addressing hierarchy, and SmartConnect DNS.

Capacity and protection sizing

  • Plan from usable capacity, not raw. Protection levels such as +1, +2, +3, +4:1, and mirrored consume capacity and determine how many node or drive failures the cluster tolerates.
  • A useful first approximation for node-level protection is usable ≈ raw × (N − protection)/N, before file-system and snapshot overhead. Add spare capacity for growth, snapshots, and rebuild headroom.
  • Performance sizing considers drive count, SSD metadata, node count, and client protocol. A single node cannot saturate a large front-end network.

Network tiers

  • Front-end: client data, typically 10/25/40/100 GbE, using redundant switches and link aggregation where supported. This network carries SMB, NFS, S3, and HDFS traffic.
  • Back-end: intra-cluster traffic on 40/100 GbE Ethernet or InfiniBand. It must be isolated, consistently cabled, and use supported switches and MTU settings.
  • Management: optional out-of-band access for IPMI and cluster administration; do not confuse it with client data.

IP hierarchy and SmartConnect

  • A groupnet contains subnets; a subnet contains IP pools; a pool supplies addresses to a SmartConnect zone.
  • SmartConnect zones give clients one DNS name. Delegate the zone in DNS to the cluster, and use SmartConnect Advanced for dynamic IP allocation and rebalancing.
  • Use VLAN tagging and consistent MTU across switches. Avoid mixing jumbo frames on one path and 1500-byte MTU on another.

Worked example

Four nodes, each with 12 × 10 TB drives, protected at +2. Raw = 4 × 12 × 10 = 480 TB. Approximate usable = 480 × (4 − 2) / 4 = 240 TB before snapshots and file-system overhead. If the customer needs 200 TB usable with 20% growth, this is too tight; add nodes or drives. Also confirm two node failures are acceptable and that back-end cabling supports all four nodes.

Traps

  • Quoting raw capacity as usable.
  • Forgetting DNS delegation for the SmartConnect zone.
  • Cabling back-end ports to the client switch or leaving back-end isolation incomplete.
  • Mixing MTU or VLAN settings across a bond.