Storage Architecture and Technologies
Domain 2: Storage Architecture and Technologies
This domain tests the mechanics behind performance, resiliency, and capacity: controller architecture, RAID, caching, thin provisioning, protocols, multipathing, and replication. Expect both calculations and “why is this design wrong?” scenarios.
Controller and data-path models
- Primera/Alletra 9000 uses a mesh-active multi-controller architecture with Persistent Ports, Virtual Domains, Priority Optimization QoS, and Peer Persistence. Its target is continuous availability, not just fast IO.
- Nimble/Alletra 6000/5000 uses CASL: flash caches writes and reads in front of sequential layout on capacity media, with triple-parity RAID and scale-out. InfoSight is part of the operating model.
- Alletra Storage MP separates compute and storage nodes in a disaggregated design for independent scaling and multi-protocol delivery.
- MSA is a simpler dual-controller array; XP8 is the high-end continuous-availability system for mainframe and open systems.
RAID, capacity, and data services
- RAID 5 tolerates one drive failure; RAID 6 tolerates two; RAID 10 mirrors and stripes. Parity RAID carries a write penalty: RAID 5 is often modeled as 4 backend IOs per write, RAID 6 as 6, RAID 10 as 2.
- Usable capacity is reduced by parity, sparing, reserved space, snapshots, and filesystem overhead. Thin provisioning does not create physical capacity.
- Snapshots are point-in-time metadata copies; clones are independent copies. Synchronous Remote Copy gives RPO 0 but adds latency; asynchronous or periodic copy gives distance at nonzero RPO.
Protocols and multipathing
FC, iSCSI, NVMe-oF (FC-NVMe and RoCEv2), SAS, NFS, and SMB each have design rules. ALUA and multipathing provide path failover and load balancing; single-fabric or single-path designs defeat redundancy.
Worked example
A volume sees 10,000 front-end IOPS at 70% read/30% write random. On RAID 6, backend IOPS = 10,000 × (0.70 + 0.30 × 6) = 25,000. On RAID 10 it is 10,000 × (0.70 + 0.30 × 2) = 13,000. RAID 6 therefore needs roughly twice the backend capability for the same front-end load, which is why write-heavy workloads often use RAID 10 or flash-optimized layouts.
Traps that cost marks
- Using RAID 5’s write penalty for RAID 6.
- Planning usable capacity without parity, sparing, and snapshot reserve.
- Assuming thin provisioning removes the need to monitor capacity.
- Claiming high availability while ignoring ALUA and multipath configuration.
- Treating synchronous replication as distance-independent.