Skip to main content
Benchmark data on a 5-node Raft cluster (c5.xlarge, us-east-1, multi-AZ 1:1:1 topology, memory hot payload backend). Client ran on a dedicated c5.xlarge instance in the same region. S2 numbers are from their published documentation (s2.dev). WarpStream numbers are from their published benchmark (6×m6in.4xl, S3 Standard, OpenMessaging Benchmark framework).

​
Latency

MetricSnakreamS2WarpStream
E2E p99 (same region)23ms<50ms~900ms
E2E p50 (same region)15ms—~500ms
Append p5013ms—~250ms
Append p9515ms—~500ms

​
Durability

AspectSnakreamS2WarpStream
Write-path storageQuorum EBS + S3 StandardS3 Express One ZoneS3 Standard
ACK semanticsMajority fsync before ACK — no data loss windowS3 Express PUT ACK — durable when S3 confirmsBatch flush to S3 — unflushed records in agent memory lost on crash
Replication3-node synchronous, majority ACKQuorum over 3 S3 Express One Zone zonal buckets—
Hot data loss requires2+ simultaneous EBS failures across AZs before repair2+ S3 Express One Zone zonal bucket failures (per-zone durability SLA unpublished)Agent crash before batch flush (default 250ms window)
Storage maturityEBS gp3 — well-understood failure modesS3 Express One Zone — no published durability SLAS3 Standard — 11 nines, proven

​
Cost (us-east-1, as of 2026-03)

Per-tenant scenario: 1K streams, 1M appends/day (500B avg), 100 GB retained, persistent connections.
Per-tenant costSnakreamS2 ExpressWarpStream
S3 cold storage (100 GB)$2.30/mo— (included)$2.30/mo
Platform / service fee$1.10/mo—~$3.11/mo (platform + agent compute)
Per-use charges (storage + transfer + ops)—$96/mo—
Per tenant$3.40/mo$96/mo~$5.41/mo