Scaling Systems: Hardware Expansion, Distributed IDs, and Elastic Capacity

Why Expand Capacity

Performance optimization has limits; for high‑traffic applications, hardware upgrades such as stronger CPUs and larger memory become necessary when software‑level optimizations like rate limiting and resource isolation are insufficient.

Expansion Strategies

Two main approaches exist: scaling an entire machine (CPU, memory, storage) or scaling individual components (memory, disk, CPU).

Whole‑Machine Scaling

Professional server vendors (IBM, Inspur, Dell, HP) provide pre‑assembled, optimized hardware, which is more reliable than DIY builds for most users.

Component Scaling

Tech‑savvy companies often purchase and assemble components themselves to reduce costs and tailor hardware to workload characteristics (CPU‑intensive, I/O‑intensive, storage‑heavy).

Key components include:

CPU : Intel, AMD, frequency, core count, etc.

Network Card : 100 Mbps → 1 Gbps → 10 Gbps

Memory : ECC verification

Disk : SCSI HDD, HHD, SATA SSD, PCI‑e SSD, NVMe SSD

AKF Splitting Principle

The AKF principle separates load across multiple machines (X‑axis) and isolates hot‑spot services for targeted scaling (Y‑axis). After business splitting, further data partitioning across regions can improve latency.

Problems After Splitting and Scaling

Data Sharing : Synchronizing data across services requires data centers or database clusters.

Interface Calls : Remote calls rely on RPC protocols such as Java RMI or Dubbo.

Persistence Snowball : Database sharding, resource isolation, and cache persistence strategies are needed.

High Concurrency : Issues like cache breakdown, penetration, and avalanche must be addressed.

Data Consistency : Distributed locks are often used to ensure consistent values across nodes.

Database Scaling: Clustering

When a single database cannot meet demand, clustering (master‑master or master‑slave) distributes reads and writes across multiple servers.

Distributed ID

Unique identifiers are essential for distributed systems. Requirements include global uniqueness, monotonic increase, and security against enumeration.

Distributed ID Requirements

Global uniqueness

Trend‑increasing order for efficient indexing

Monotonic increase for versioning

Security to prevent predictable IDs

Distributed ID Generation Strategies

Common methods include UUID, Snowflake, Redis, Zookeeper, etc. This article focuses on UUID and Snowflake.

UUID Generation

UUIDs are 128‑bit identifiers represented as 36‑character strings (e.g., 550e8400-e29b-41d4-a716-446655440000). Advantages: high performance, no network overhead. Disadvantages: large size, potential MAC address leakage, and poor suitability as primary keys in databases due to length and randomness.

Snowflake Algorithm

Snowflake uses a 64‑bit structure divided into timestamp, machine ID, and sequence fields, enabling trend‑increasing IDs with high QPS. Advantages: time‑ordered IDs, no database dependency, flexible bit allocation. Disadvantages: reliance on accurate system clocks; clock rollback can cause duplicate IDs.

Elastic Scaling

Elastic scaling automatically adjusts resources based on scheduled demand, reducing idle capacity. Challenges include weak elasticity of virtual machines, multi‑department coordination, and high IT costs from over‑provisioning.