Planning and Designing Storage Solutions
Domain 3: Planning and Designing Storage Solutions
This domain tests requirements gathering and the translation of business constraints into a defensible storage design: capacity, performance, availability, network, replication, backup, and growth. A correct answer is traceable from requirement to design decision.
Requirements to design inputs
- Capture IOPS, throughput, latency target, read/write mix, IO size, working set, usable capacity, growth rate, data-reduction assumptions, snapshot retention, replication, backup window, RPO, RTO, compliance, and budget/consumption model.
- Availability tier drives redundancy: dual fabrics, dual power, multipathing, N+1 controllers, site separation, quorum witness, and tested runbooks. A 100% guarantee tier is a design commitment, not a marketing label.
- Capacity math must include RAID efficiency, sparing, reserved space, snapshot and clone usage, filesystem overhead, and free-space headroom. Performance sizing must include backend write penalty and replication load.
Network and host design
- FC: single-initiator/single-target zoning by WWN, separate VSANs, ISL oversubscription limits, and consistent fan-in.
- iSCSI: dedicated VLANs, jumbo-frame consistency end to end, port binding, CHAP where required, and redundant paths.
- NVMe-oF: confirm adapter, switch, and OS support; RoCEv2 requires lossless configuration.
- Multipathing: VMware NMP with an appropriate PSP, Windows MPIO, Linux DM-MP, and correct ALUA settings.
Replication and DR
Synchronous replication gives RPO 0 but is latency- and distance-sensitive. Asynchronous or periodic replication supports longer distances with nonzero RPO. Peer Persistence needs synchronous Remote Copy plus a Quorum Witness. Always design the failover/failback runbook and test it.
Worked example
A customer needs 100 TB usable today, expects 30% growth, wants 20% free space, and uses RAID 6 with 8+2 parity and 10% sparing. Required protected capacity = 100 × 1.30 × 1.20 = 156 TB. Raw capacity = 156 ÷ 0.80 ÷ 0.90 ≈ 216.7 TB. The design must then check that the chosen model can deliver the IOPS and latency targets at that capacity, not just the terabytes.
Traps that cost marks
- Sizing capacity only and ignoring performance and replication bandwidth.
- Forgetting RAID, sparing, snapshot, and free-space overheads.
- Assuming synchronous replication works at any distance or latency.
- Designing a single fabric or single path and calling it highly available.
- Skipping a validation plan or PoC acceptance criteria.