Scaling Strategies, Hardware Expansion, and Distributed ID Generation in Backend Systems

Why expand capacity: performance optimization has limits; when traffic grows beyond what software tuning can handle, stronger CPUs, more memory, or additional servers are required, similar to adding more windows in a busy cafeteria.

Scaling strategies: two main approaches – whole‑machine hardware expansion (CPU, memory, storage) and component‑level upgrades (CPU, NIC, memory, disks). Whole‑machine upgrades benefit from professional vendors, while component upgrades allow custom tailoring for compute‑ or I/O‑intensive workloads.

AKF splitting principle: X‑axis scaling distributes requests across multiple machines (master‑master, master‑slave clusters) but introduces data synchronization challenges; Y‑axis splitting isolates hot business logic for targeted scaling; further Y‑axis splitting can replicate data across geographically distributed data centers.

Problems after splitting include data sharing, RPC interface calls (Java RMI, Dubbo), database sharding and cache avalanche, high concurrency, and data consistency, often mitigated with distributed locks.

Database scaling – clustering: distinguishes distributed systems (task parallelism) from clusters (increasing operation throughput). Database clusters (master‑master or master‑slave) spread read/write load across multiple nodes.

Distributed ID requirements: global uniqueness, monotonic or trend‑increasing order, and security (non‑predictable IDs). High availability (5‑9 or 6‑9 uptime) and low latency are also essential.

ID generation strategies: UUID (e.g., 550e8400-e29b-41d4-a716-446655440000 ) offers high performance but is long, may expose MAC addresses, and can degrade index performance; Snowflake uses a 64‑bit structure (timestamp, machine ID, sequence) to produce trend‑increasing, high‑throughput IDs, though it depends on accurate clocks.

Elastic scaling: automatically adjusts resources based on scheduled demand, reducing idle capacity and IT costs, but faces challenges such as slow VM provisioning and high overhead.