Designing API Performance Optimization

1. Designing Response Compression

Algorithm	Use	Notes
gzip	Universal default	Good ratio, low CPU
brotli	HTTPS responses	~20% better than gzip; modern browsers
zstd	Internal RPC	Fast, tunable
Threshold	Compress >1KB only	Avoid CPU on small bodies

2. Designing API Caching Strategy

Header	Effect
Cache-Control: max-age=N	Client caches N seconds
s-maxage	Shared cache (CDN) overrides max-age
stale-while-revalidate	Serve stale, refresh in bg
ETag / If-None-Match	304 Not Modified
Vary	Cache key dimensions (Accept-Encoding, Accept-Language)

3. Designing Efficient Payload Structures

Practice	Detail
Compact JSON	Drop nulls, short keys for hot endpoints
Binary formats	Protobuf, MessagePack, CBOR for internal
Streaming	NDJSON / chunked transfer for large lists
Pagination	Don't return full collections

4. Designing Field Selection

Approach	Example
Sparse fieldsets (JSON:API)	?fields[user]=name,email
GraphQL	Client picks exact fields
Google APIs style	?fields=name,email,address(city)

5. Using HTTP/2 and HTTP/3

Version	Wins	Notes
HTTP/1.1	—	HoL blocking, 6 conns/host
HTTP/2	Multiplex over 1 TCP, header compression (HPACK), server push (deprecated)	Single TCP HoL
HTTP/3 (QUIC)	UDP, 0-RTT, no HoL across streams, faster handshake	Better mobile networks

6. Designing Connection Keep-Alive Strategy

Setting	Recommendation
Keep-Alive timeout	60–120s
Max requests/conn	1000+ (avoid premature close)
Connection pool (client)	Reuse across requests
LB idle timeout	Match upstream

7. Designing Request Batching

Strategy	Detail
Client batch	Group N requests into 1 (e.g., GraphQL batch)
Server-side coalescing	DataLoader, micro-batching
Time window	10–50ms batch window
Trade-off	Latency for throughput

8. Designing Lazy Loading Strategy

Scope	Approach
Field-level	Resolve on demand
Pagination	Load chunks on scroll
Sub-resource	HATEOAS link, follow when needed
Defer / stream (GraphQL)	Send slow fields after first byte

9. Designing API Response Time Optimization

Lever	Impact
Cache layer	10–100× for hits
Reduce DB queries (N+1)	Often 5–50×
Index tuning	Bring queries to ms
Async non-critical work	Return faster
Co-locate services	Eliminate cross-AZ hops

10. Designing Performance Budgets

Budget Type	Example
Latency	p95 < 200ms for /api/*
Payload size	< 50KB gzipped
Time to first byte	< 100ms
Error budget	< 0.1% 5xx
Enforcement	CI perf tests fail build

11. Designing Payload Compression

Layer	Detail
Transport	gzip/brotli/zstd
Serialization	Protobuf vs JSON: 3–10× smaller
Image/video	WebP/AVIF, AV1, modern codecs
Field-level	Diff/patch (JSON-Patch) for updates

12. Designing Network Optimization

Technique	Detail
CDN / PoPs	Reduce RTT to origin
TCP BBR / TLS 1.3	Faster congestion + handshake
Connection prewarm	Keep N idle conns ready
Region pinning	Same-AZ where possible
Anycast	Lower client RTT globally