# Tiered storage won't fix Kafka #reading-list **URL:** https://www.warpstream.com/blog/tiered-storage-wont-fix-kafka ## Summary - Tiered Storage is an idea in Kafka and related systems that allows brokers to ship log segments off to cloud object storage (S3). The author argues that Tiered Storage fails at all of its primary goals - **Goal 1: reducing cost**. The author lays out an argument that inter-zone networking, not disks, are actually the majority of cost associated with running Kafka. Because tiered storage only occurs after a new record has been fully ACKed, it does nothing to reduce network costs. It may reduce disk costs but not substantially. - Goal 2: operational flexibility. Doing large replays or batch jobs requires reading the object store data. Existing tiered storage fails at this - Segments in S3 are still managed by the brokers, so reading directly is not safe as there is no snapshot isolation - Large replays usually require reading full segments from the object store to broker disk before serving queries, which can be 100x slower than a local read