Evaluating In-Memory Database Performance on the HaiGuang CPU: Challenges, Requirements, and Application Scenarios

With the rapid development of cloud computing and internet technologies, the explosion of data volume and diverse user scenarios have placed unprecedented pressure on traditional databases and raised the demand for higher CPU performance. In the context of domestic technology replacement, new database solutions based on Chinese chips have attracted significant attention.

Recently, a project successfully ran an in‑memory database on a HaiGuang CPU, providing a concrete case study of how domestic processors can support high‑performance memory‑centric workloads.

Challenges faced by traditional databases

Data scale growth: Explosive data growth strains storage, backup, and recovery, while real‑time access requirements create performance bottlenecks.

Data diversity: The rise of semi‑structured and unstructured data (social media, logs, multimedia) challenges conventional relational systems.

Availability and scalability: Modern applications need highly available, easily scalable databases, which many legacy systems cannot provide.

To address these issues, enterprises are turning to new solutions such as in‑memory, distributed, and NoSQL databases, which offer higher performance, better scalability, and lower cost, often delivered as Database‑as‑a‑Service (DBaaS) in cloud environments.

What is an in‑memory database?

An in‑memory database stores the primary data set in RAM rather than on disk, enabling ultra‑fast data access. Its key characteristics include resident primary copies in memory, a redesign of storage architecture away from disk‑centric models, performance gains often exceeding tenfold, and suitability for real‑time analytics and transaction processing.

CPU requirements for in‑memory databases

In‑memory databases demand powerful CPUs to handle massive data operations and low‑latency queries. High core counts, high clock speeds, large last‑level caches, rich instruction sets (SSE, AVX, prefetch, atomic operations), and robust security features (hardware encryption, ECC support) are critical.

HaiGuang CPU evaluation

The HaiGuang 7300 offers 32 physical cores and 64 threads, enabling parallel processing of batch queries and transactions, markedly improving query performance.

Its base clock of ≥2.9 GHz provides faster instruction execution and data processing for memory‑resident workloads.

A 64 MB L3 cache reduces memory accesses, boosting cache‑hit rates essential for high‑throughput in‑memory operations.

The processor’s rich instruction set—including SSE*, AVX*, prefetch, and atomic instructions (MFENCE, LFENCE, SFENCE)—directly benefits in‑memory database workloads.

Built‑in security accelerators support national cryptographic standards (SM2/3/4), enabling fast encryption/decryption without additional hardware, and the impact on latency is minimal.

Support for ECC‑enabled CPUs and memory modules, along with Reed‑Solomon (RS(140,128)) error‑correction, ensures data integrity and reliability.

Application scenarios for in‑memory databases

Real‑time applications such as online advertising and financial trading, where millisecond‑level latency is essential.

Big‑data processing and real‑time analytics, leveraging high concurrency and fast data access.

Scientific computing and risk simulation, benefiting from rapid in‑memory data manipulation.

Telecommunications services (core network, CRM, precise marketing), where fast data handling supports complex, real‑time business logic.

Overall, the HaiGuang processor’s performance, security, and reliability features make it a strong candidate for powering in‑memory database deployments across a wide range of demanding workloads.