Benchmark lower-copy Redis protocol alternatives for large payloads

[lan17]

Priority

P3 — Low — later or demand-driven enhancement.

Problem

Reads perform GET followed by string.sub in Lua, and writes concatenate the entire frame before SET. Preliminary measurements show overhead grows sharply with payload size.

Evidence:

• Whole-value read and slice
• Whole-frame concatenation

Acceptance criteria

• Reproduce tracked and untracked reads/writes at 100 B, 10 KB, 100 KB, and 1 MB against raw Redis operations.
• Test the Redis 6.2 support floor and current supported Valkey.
• Evaluate lower-copy alternatives while preserving binary serializers, ordering, targeted invalidation, and cluster routing.
• Record Redis CPU, throughput, latency, wire round trips, and temporary memory.
• Produce an approved design decision before changing the wire protocol.
• Keep immediate payload guardrails in their separate issue.