การลดแบนด์วิดท์ด้วยการบีบอัดข้อความ

Real-time apps can push thousands of frames per second. On mobile and high-fan-out systems, raw bytes on the wire become a real cost in latency, data charges, and egress bills.

WebSocket defines an extension, permessage-deflate, that compresses each message with DEFLATE (the same algorithm as gzip) before it is sent and decompresses it on arrival.

Compression is negotiated via the Sec-WebSocket-Extensions header during the upgrade. If both peers agree, frames are compressed transparently afterward.

Compression helps most with text and JSON, which are highly repetitive. A 4 KB JSON payload can shrink to under 1 KB.

Already-compressed data (images, video, gzip) barely shrinks and just wastes CPU.

Compression is not free. Every message costs CPU on both sides. Under very high message rates the CPU spent compressing can outweigh the bandwidth saved.

DEFLATE keeps a sliding-window dictionary per connection (context takeover). With many connections this is significant memory. Disabling takeover saves RAM but lowers the compression ratio.

Spring delegates to the servlet container. On Tomcat you tune the WebSocket compression via container properties or a customizer.

Compressing tiny frames adds overhead with little gain. Many stacks skip compression below a threshold (e.g. messages under 256 bytes). Keep this in mind when designing payload sizes.

Before reaching for transport compression, shrink the payload itself:

• Send deltas, not full snapshots
• Use short field names or a binary format (Protobuf, MessagePack)
• Batch many small updates into one frame

Always measure. Compare bytes-on-wire and CPU before and after enabling compression for your real traffic mix; do not assume it is always a win.

Rules of thumb:

• High-volume JSON, bandwidth-bound → enable permessage-deflate
• Many idle connections, memory-bound → disable context takeover
• CPU-bound, small frames → prefer application-level payload shrinking

Test your compression knowledge.

You optimized bandwidth:

• permessage-deflate compresses each message, negotiated in the handshake
• Best for repetitive text/JSON; useless for already-compressed data
• It trades CPU and per-connection memory (context takeover) for fewer bytes
• Application-level deltas, binary formats, and batching are complementary
• Always measure before and after for your real traffic