Key Architectural Features and Technical Considerations of Flash Storage Systems

Flash storage is characterized by stability, low latency, and high IOPS; performance evaluation often considers that 90% of I/O falls within a specified latency range, emphasizing that performance is a linear range rather than a single point. Enhancements like inline deduplication, compression, and thin provisioning improve performance and data protection while reducing costs and extending SSD lifespan.

Flash Architecture: The scale‑out capability is essential for handling concurrent access and increasing performance capacity. Modern flash arrays such as XtremIO and SolidFire support multiple controllers (up to 16 and 100 respectively) to provide this horizontal scalability.

Controller Symmetric A/A Capability: For workloads like OLTP, traditional A/P or ALUA arrays incur I/O reset during controller failover, and the system CPU often becomes a bottleneck. Only an active‑active (A/A) symmetric architecture, which avoids I/O reset and balances performance, can meet these demands. While many traditional storage systems (EMC VMAX/VNX, HPE 3PAR, HDS USP/VSP) support A/A, many flash products still use A/P designs.

Metadata Management: Flash systems focus on exploiting SSD random‑access performance without relying on prefetch or I/O aggregation techniques used in HDDs. Consequently, design starts from metadata (system metadata, deduplication/compression fingerprints, FTL mapping), I/O processing strategies, garbage collection, and wear‑leveling. A two‑layer metadata architecture maps LBA → block ID → physical location, simplifying deduplication and compression implementation.

Global FTL (GFTL) Functionality: GFTL enables flash arrays to cooperate with SSD controllers (requiring open SSD interfaces) to perform advanced optimizations such as global load balancing, garbage collection, unified address/data management, and full‑stripe writes. It can also record deduplication/compression databases, implement RAID‑like functions, and mitigate write‑hole issues. Although many SSD vendors do not yet support GFTL, it is a powerful tool for improving flash efficiency.

Deduplication Features: Deduplication is a core flash capability, available as inline or post‑line, with inline offering true value. Common algorithms include hash‑based (e.g., SHA‑1, SHA‑256) and byte‑by‑byte comparison, though the latter heavily impacts performance and is rarely used. Strong hash algorithms have extremely low collision probabilities, making them suitable for most scenarios.

I/O Basic Flow: Data from the host is sent to the flash array controller. For non‑A/A designs, the controller first determines LUN ownership, then caches data, splits it into appropriate block sizes, passes it to value‑added modules (deduplication/compression), generates fingerprints, and finally writes data to the appropriate controller, allocating stripes and recording physical addresses.

Block Wear‑Leveling: To distribute data evenly across SSD blocks and avoid hotspots, flash systems implement both dynamic and static wear‑leveling. Dynamic wear‑leveling spreads writes triggered by host updates across all blocks, while static wear‑leveling moves data from cold blocks to hot ones, balancing block usage over time.

SSD Power‑Loss Protection: To prevent loss of cached data during power failures, SSDs incorporate voltage detection modules that trigger a power‑loss protection sequence. When voltage drops below a threshold, the controller initiates a data flush to flash cells, often using a super‑capacitor as backup power to ensure data integrity.

For more detailed information, please follow the provided QR code or click the original link to access additional ICT resources.