Emerging Trends and Challenges in Cloud and Internet Storage: Towards a New Disaggregated Compute‑Storage Architecture

From the perspective of cloud and internet business scenarios, the storage domain primarily adopts a server‑deployed distributed storage service model, which faces several challenges: mismatched data retention and server update cycles, difficulty balancing performance reliability with resource utilization, low space efficiency of traditional replication, and the growing "datacenter tax" that reduces efficiency for data‑intensive applications.

To address these issues, recent hardware trends such as Ethernet Bunch of Flash (EBOF), high‑performance NVMe over Fabric (NoF), Data Processing Units (DPUs), and advanced interconnects like CXL are rapidly evolving, providing the technical foundation for a new disaggregated compute‑storage architecture.

The proposed architecture features a thorough decoupling of compute and storage resources, forming independent hardware pools (CPU, memory, HDD/SSD) that can be scaled and shared flexibly; finer‑grained task partitioning where specialized accelerators (DPUs, GPUs, FPGAs) replace general‑purpose CPUs for data‑heavy workloads; and a diversified network protocol stack (CXL, NoF, IP, Fibre Channel) that delivers sub‑microsecond latency and high‑throughput data paths.

Key characteristics of the new architecture include diskless servers with remote storage pools, multi‑protocol networking (CXL + NoF + IP), dedicated data processors for tasks such as erasure coding and encryption, and ultra‑dense storage systems that combine traditional RAID/EC with modern high‑density flash modules.

By reorganizing storage modules (EBOF, EBOM, EBOD), compute modules (DPUs, specialized accelerators), and high‑throughput data buses, the architecture aims to solve the earlier listed pain points, offering improved resource efficiency, higher reliability, and better performance for cloud and internet workloads.