HPE ASE Storage Architect cheat sheet

Capacity sizing formula: usable capacity = raw capacity × (1 − parity/rebuild overhead) × (1 − reserved spare %); RAID 6 with 8 data disks loses 25% to parity.

RAID decision rule: RAID 1/10 for write-heavy low-latency workloads, RAID 5 for read-heavy sequential, RAID 6 when rebuild window or disk size exceeds acceptable risk, RAID 0 never for protected data.

IOPS sizing: total IOPS = (drive IOPS × drive count) / read-write amplification; SSD random read ≈ 10k–100k+ IOPS per drive, 15K SAS HDD ≈ 180 IOPS, 7.2K NL-SAS ≈ 75 IOPS.

Latency budget rule: if application SLA is under 1 ms, use all-flash with NVMe; 1–5 ms fits hybrid SSD tier; above 10 ms tolerates HDD-backed capacity tiers.

Throughput formula: sequential MB/s ≈ IOPS × block size (KB) / 1024; verify both IOPS and MB/s against workload profile, not just one.

Protocol selection: FC for deterministic low-latency block on SAN fabrics, iSCSI for cost-sensitive block over Ethernet, NFS/SMB for file sharing, NVMe-oF (FC or TCP) for sub-100 µs block.

Thin provisioning overcommit ratio = provisioned capacity / physical capacity; keep below 2:1 unless monitoring alerts fire at 70–80% pool consumption.

3-2-1 protection rule for design: at least 3 copies of data, on 2 different media or systems, with 1 copy offsite or offline.

Erasure coding rule: k data + m parity fragments tolerate m failures; overhead = m/k; use when m/k is lower than RAID parity overhead at scale (e.g., 8+2 = 25%).

FC zoning rule: single-initiator, single-target zones by WWPN; never zone by WWNN for host-to-array traffic, and keep the same host HBA in one zone per fabric.

Multipath requirement: each host needs at least two independent paths (two HBAs, two fabrics, two array ports); ALUA reports active/optimized vs. active/non-optimized paths.

iSCSI best practice: separate storage VLAN, jumbo frames (MTU 9000) end-to-end or not at all, CHAP for authentication, and MPIO with at least two sessions.

LUN masking order: create host object with WWPN/IQN, create volume, map volume to host set, then rescan host bus; verify with OS-level multipath tools.

Array initialization sequence: physical install → management IP → firmware baseline → disk pool/RAID group → volumes → host mapping → multipath → filesystem → application validation.

NVMe-oF deployment: requires lossless or congestion-controlled fabric (PFC/ECN for RoCEv2), discovery controller configuration, and host namespace scan after connect.

Firmware rule: match array controller, drive, and host HBA/multipath driver versions to the vendor interoperability matrix before production cutover.

Snapshot/clone provisioning: reserve capacity sized to expected change rate × retention window; monitor snapshot pool to prevent forced deletion of oldest snapshots.

Migration cutover: baseline performance, replicate, quiesce application, final sync, remap host paths, validate, then decommission source after rollback window.

Performance troubleshooting order: check host queue depth and multipath path state, then fabric/network errors, then array port utilization, then backend disk latency.

Utilization thresholds: array port above 70% sustained, backend disk above 80% busy, or pool above 80% full are typical escalation triggers.

Queue depth rule: optimal host queue depth ≈ (IOPS per path × average latency in seconds); excessive depth raises latency without raising throughput.

Alerting design: separate warning (predictive) from critical (failed component) tiers; deduplicate by object so one failed disk does not flood the ticketing system.

Capacity reporting: track subscribed, used, and snapshot overhead separately; thin pool growth rate trends predict exhaustion date better than a single percentage.

Change management: every firmware or zoning change needs a rollback plan, a maintenance window, and post-change validation of multipath and application latency.

Health monitoring: SNMP traps or REST API polling for drive predictive failure, fan/power supply state, replication lag, and certificate expiry.

Replication operations: monitor RPO lag, not just link state; a link that is up but lagging breaches RPO silently.

Log collection: capture array event logs, host multipath logs, and fabric logs with synchronized NTP timestamps before escalation.

RPO defines maximum tolerable data loss; RTO defines maximum tolerable downtime. Synchronous replication gives near-zero RPO; asynchronous gives RPO equal to replication interval.

Snapshot is not a backup: it shares the source volume's failure domain; a backup requires an independent copy on separate media or site.

Immutable/air-gapped copies defend against ransomware encryption; retention lock prevents deletion or modification for a defined period.

Encryption rule: data-at-rest encryption protects against physical drive theft, not against a compromised host with mounted volumes; use host or application encryption for that.

Key management: external key manager with at least two key servers; losing all key copies makes encrypted data unrecoverable, including backups.

Replication consistency: crash-consistent copies reflect an abrupt power loss; application-consistent copies use quiesce or VSS snapshots before replication.

RBAC and auditing: separate array admin, storage operator, and read-only roles; enable audit logging of configuration changes and send logs off-array.

Backup verification: run scheduled restore tests, not just backup job success; an unverified backup has unknown RPO and RTO.

Secure transport: use FC zoning plus fabric authentication (e.g., DH-CHAP) and IPsec or TLS for replication links crossing untrusted networks.