Write Lifecycle Routing Simulator
Inspecting: AutoMQ Fast-WAL Loop (FSx NetApp ONTAP)
The 5 Diskless Latency Contracts (KIP-1150 Spectrum)
Select contract to simulate routing loop
Monthly Cloud Egress & Storage TCO Modeling
Continuous Ingress: 20 MB/s (51.84 TB/mo)
Classic 3-AZ Replicated ISR Kafka
Cross-AZ Ingress/Egress Replicas
103.68 TB / mo
Cloud Cross-AZ Egress Penalty
$2,073.60 / mo
EBS gp3 Storage Tier (3 Replicas)
$1,244.16 / mo
Baseline Infrastructure TCO
$3,317.76 / mo
Selected Architecture: AutoMQ Fast WAL
Cross-AZ Replication Overhead
0 TB (Eliminated)
Cross-AZ Egress Cost
$0.00 / mo
Object Storage & WAL Fees
$1,192.32 / mo
Diskless Monthly TCO
$1,192.32 / mo
Cloud Bill Delta:
-64.1% Cost Reduction
KIP-1150 ("Diskless Topics") Architecture Specification
Accepted March 2, 2026
The Core Conflict: Apache Kafka's traditional partition model couples compute (broker request dispatch), caching (page cache), and durability (local persistent disk with multi-AZ replication). KIP-1150 decouples storage by delegating commit acknowledgments to remote tiered storage backends without mandatory cross-broker TCP replication.
- Why the 270x Latency Spread Exists: A fast write-ahead log (like AutoMQ's NetApp ONTAP/EBS pooling) allows
acks=allto return in 12.87ms P99. Conversely, writing unbuffered directly to S3 or GCS multipart batches requires accumulating 500ms to 3,500ms of payload chunks before HTTP PUT confirmation. - The Egress Trap: Moving logs straight to S3 cuts cloud provider inter-AZ replication fees to zero. However, if your consumer workloads require immediate real-time read-after-write semantics, brokers must fetch from S3 (introducing GET latency and S3 egress charges) unless cached in memory.
- SLA Mismatch Warning: KIP-1150 is ideal for 20 MB/s distributed application tracing, CDC mirrors, and audit streams. It is an architectural anti-pattern for sub-50ms financial clearing or synchronous checkout pipelines unless deployed with a low-latency WAL overlay.