Let's deep dive into the shared-storage architecture of AutoMQ(github.com/AutoMQ/automq): 1. The classic Kafka architecture, known as shared nothing, requires each broker to have a TiB-level local disk and 3 replicas. However, this setup can lead to high costs and scalability challenges in the cloud-native environment. 2. Tiered storage has indeed reduced storage costs, but the need for expensive replication and large local disks still remains. Yes, tiered storage won't fix Kafka. 3. AutoMQ's shared storage architecture utilizes EBS and S3 for shared storage. EBS functions as the WAL for recovery, while S3 serves as the primary storage. Each broker only needs a small EBS volume, making it a cost-effective solution. This design encompasses all the benefits touted by the confluent freight cluster or WarpStream, while maintaining crucial low latency for stream processing.