Measured - byte-exact - 2026

Pedulli vs Snappy

measured head-to-head - structured data classes - byte-exact SHA-256 verified

TL;DR - honest
Google Snappy was designed to maximize throughput at the cost of ratio. Pedulli is a best-of-N racer - it races xz, zstd, brotli and your data's SRD math, keeps the smallest verified output, with a +1 byte never-worse floor -> never larger than the best standard codec (worst case +1 byte). It wins outright on structured data and ties the best codec on already-optimal/random data; Snappy is faster on raw streaming binary. They serve different jobs - Snappy in hot read paths (Cassandra, HBase, Kafka), Pedulli in storage tiers where bytes compound.

The measured table

All numbers measured on this server, roundtrip-verified SHA-256 byte-exact. Proofs available on request.

Input	Snappy	Pedulli (best-of-N)	Δ
1 MiB of zeros	1,028 B	13 B	much smaller (redundant input)
JSON 31 KB	9,124 B	1,265 B	-86%
HTML 161 KB	85,432 B	38,757 B	-54.6%
Apache logs 3.5 MB	612 KB	217 KB	-64.5%
MP4 master 10 MB	10,234,621 B	9,430,108 B	-805 KB
Random bytes 1 MB	1,048,627 B (+51 B)	1,048,577 B (+1 B)	50 B less overhead

What Snappy does better (honest)

Pure throughput. Hot-path optimized for fast compress and very fast decompress on a single core.
Hadoop / Kafka / Cassandra ecosystem default. Drop-in for those stacks.
Google-vetted, battle-tested at very large scale internally.

What Pedulli does that Snappy does not

+1 byte never-worse floor vs Snappy's +51 on random data.
Best-of-N racing. Races xz, zstd, brotli and your data's SRD math, keeps the smallest verified output - never larger than the best standard codec (worst case +1 byte). Far tighter on every structured class measured.
Web service + REST API + chunked upload.
Zero data retention. GDPR by design, EU-sovereign.

When to use which

Use Snappy in hot read paths where decompression latency dominates (column stores, RPC compression, hot caches). Use Pedulli when bytes-stored matters: cold storage, log archives, video masters, JSON archival - plus the per-file +1 byte never-worse floor.

Run the free trial Pricing / API All benchmarks

More comparisons: lz4 - lzop - LZO - gzip