Modern Storage System Architecture: Components and How It Works
Modern storage systems (SAN, NAS, object storage) are sophisticated engineering solutions that combine hardware and software components to deliver high performance, reliability, and scalability. Understanding storage architecture helps IT professionals design infrastructure correctly, select appropriate hardware, and optimise workloads. In this article, we break down the key components of a storage system: controllers, disk shelves, cache, access protocols (FC, iSCSI, NVMe‑oF), and data protection technologies — RAID and erasure coding. We pay special attention to all‑flash arrays, which are becoming the standard for high‑performance environments.
1. Storage Controllers – The Brain of the System
Storage controllers are the central processing modules that manage all I/O operations, distribute data across drives, handle caching, and provide connectivity to servers. Modern storage systems use two or more controllers in an active‑active configuration for high availability.
Controller functions:
- Processing I/O requests from servers.
- Managing read/write cache.
- Implementing RAID or erasure coding.
- Load balancing between disk shelves.
- Ensuring high availability (failover).
Controllers can be hardware‑based (dedicated boards with their own processors) or software‑based (SDS — software‑defined storage). In enterprise all‑flash arrays like Huawei OceanStor Dorado 3000 or Dell PowerStore 1200T, high‑performance controllers with dedicated compression and deduplication chips are used.
Selection tip
When choosing a storage system, pay attention to the number and type of controllers. For mission‑critical systems, an active‑active dual‑controller configuration is mandatory. Also evaluate the computational power of the controllers — it directly affects maximum IOPS performance.
2. Disk Shelves and Drives
Disk shelves are modules that house physical drives (HDD, SSD, NVMe). They connect to controllers via SAS, SATA, or NVMe interfaces. Modern shelves support hot‑swap drive replacement without powering down the system.
- HDD – high capacity but slow; used for archives and cold data.
- SSD (SATA/SAS) – fast but more expensive; suitable for mixed workloads.
- NVMe – ultra‑fast with minimal latency; designed for all‑flash arrays.
Examples: Dell PowerVault ME5024 supports up to 24 drives in 2.5″ or 3.5″ form factors, while Lenovo ThinkSystem DE4000H can combine up to 192 drives in a hybrid configuration.
3. Cache Memory
Cache is high‑speed memory (usually DRAM or NVRAM) located on the controllers. It acts as a buffer between fast server requests and slower disk media. Cache is split into:
- Read cache – stores frequently accessed data to accelerate reads.
- Write cache – buffers write operations to smooth out peak loads and reduce latency.
In all‑flash arrays, cache is used to accelerate operations and also for inline deduplication and compression. For example, Huawei Dorado 5000 uses intelligent caching with machine‑learning algorithms for data prefetching.
4. Access Protocols: FC, iSCSI, NVMe‑oF
The choice of protocol determines how servers connect to the storage system and affects performance, latency, and infrastructure cost.
Fibre Channel (FC)
A high‑speed protocol based on dedicated optical networks. Delivers low latency and high throughput (up to 64 Gbit/s). Used in enterprise SANs. Requires FC switches and HBA adapters.
iSCSI
Encapsulates SCSI commands in TCP/IP packets. Operates over standard Ethernet, lowering cost but adding latency due to TCP/IP overhead. Suitable for mid‑sized enterprises with budget constraints.
NVMe‑oF (NVMe over Fabrics)
The most modern protocol, extending NVMe over networks (Ethernet, InfiniBand, FC). Delivers microsecond‑scale latency and throughput comparable to local NVMe. Ideal for all‑flash arrays and high‑performance computing. Requires RDMA‑capable network adapters (RoCE, iWARP).
Our data centre switches, such as Cisco Nexus 9300‑GX2, support high‑speed connections for NVMe‑oF.
Which protocol to choose?
FC – for the most demanding enterprise environments with existing FC infrastructure. iSCSI – for smaller companies looking to save on networking. NVMe‑oF – for new projects where performance and low latency are critical, especially for AI, Big Data, and high‑throughput databases.
5. Data Protection: RAID and Erasure Coding
Both technologies provide data redundancy to protect against drive failures, but they work differently.
RAID (Redundant Array of Independent Disks)
Hardware‑ or software‑based method of combining drives into arrays with various redundancy levels (RAID 0, 1, 5, 6, 10). RAID 5 and 6 use parity to reconstruct data when one or two drives fail. RAID is widely used in DAS, NAS, and entry‑level SAN.
Erasure Coding
A more modern method used in object and distributed storage systems. Data is split into data chunks and parity chunks, which are distributed across different nodes. This allows recovery from multiple node failures simultaneously. Erasure coding provides better storage efficiency compared to replication and is often used in scale‑out systems such as Huawei OceanStor Pacific 9550.
6. Architectural Approaches: Scale‑Up vs Scale‑Out
Understanding these approaches helps choose the right storage system for business growth.
- Scale‑Up (vertical scaling) – increasing capacity and performance by adding disk shelves or more powerful controllers to an existing system. Suitable for predictable workloads. Examples: traditional SAN and NAS.
- Scale‑Out (horizontal scaling) – adding new nodes (controllers with drives) to a cluster, which increases both capacity and performance linearly. Ideal for big data, cloud environments, and unstructured data. Example: Huawei Pacific 9550 and other object storage systems.
7. Modern All‑Flash Architectures
All‑flash arrays are built entirely on SSDs/NVMe. They deliver minimal latency (below 0.1 ms) and high throughput, which is critical for OLTP, virtualisation, and AI workloads. Key features:
- NVMe drives for maximum speed.
- Built‑in deduplication and compression to save space.
- Intelligent caching and automatic optimisation.
- Support for NVMe‑oF for low‑latency network access.
We offer all‑flash storage systems from Huawei and Dell, such as OceanStor Dorado 5000 and PowerStore 1200T, which meet the highest demands.
8. Frequently Asked Questions (FAQ)
What is an active‑active controller and why is it needed?
What is the difference between RAID and erasure coding?
Which protocol should I choose for an all‑flash array?
How does cache affect storage performance?
Build a reliable storage infrastructure with GHI‑Servers
We supply storage systems of all types — from classic SAN to modern all‑flash and object storage. Direct sourcing from China, full warranty, engineering audit, and support.
Get a commercial offerIn the next article, we will dive deep into all‑flash storage and its business benefits. Stay tuned!