Latency budget на собеседовании системного аналитика

Что такое latency budget

Total latency SLA broken down между components.

Total p99 budget: 500ms.
  - Auth check: 50ms.
  - DB query: 200ms.
  - Business logic: 50ms.
  - External API call: 150ms.
  - Serialization / network: 50ms.

Каждый owner свою долю.

Breakdown по components

Trace request through system. Allocate budget proportionally к expected work.

Caveats:

• Network roundtrips often underestimated.
• Cache misses должны быть в budget.
• Retries — extra latency.

Sequential vs parallel

Sequential calls — sum latencies.

A → B (50ms) → C (100ms) → D (50ms) = 200ms.

Parallel — max.

A → [B, C, D] in parallel → max(50, 100, 50) = 100ms.

Parallelize where possible.

Slow chains

Long sync RPC chains compound.

Service A → B → C → D → E
50ms each → 250ms total.

Plus retries / failures.

Solution. Async events для chains, или aggregate в parallel.

Optimization strategies

Caching. Avoid hot path computation.

Async. Move work из synchronous path.

Batching. Combine multiple ops в one round-trip.

Connection pooling. Reuse connections.

Compression. Less network bytes.

Edge / CDN. Closer к user.

Read replicas. Distribute read load.

Pre-computation. Materialized views, pre-aggregated tables.

Связанные темы

• SLA SLO SLI для SA
• Cache strategies для SA
• Capacity planning для SA
• Circuit Breaker для SA
• Подготовка к собесу системного аналитика

FAQ

Это официальная информация?

Нет. Статья основана на индустриальных performance practices.

Тренируйте системный анализ — откройте тренажёр с 1500+ вопросами для собесов.