What Five P99 CONF Talks Reveal About Database Trade-Offs
The New Stack, Friday, October 30th, 2026
ScyllaDB-sponsored New Stack recap of five P99 CONF 2026 talks on how databases trade off read, write and space amplification.
This post, sponsored by ScyllaDB and written by Doug Sillars, covers five sessions from the October 2026 P99 CONF on low-latency systems.
Almog Gavra explains LSM trees and SlateDB for object storage, and ScyllaDB engineers present Logstor, an opt-in engine that cuts read amplification at the cost of RAM and write overhead.
AWS's Madelyn Olson shows how Valkey raised a single-node ceiling about eightfold with I/O threads and added atomic slot migration.
Turbopuffer describes searching 100 billion vectors at 200 ms p99, and TigerBeetle explains building its own object storage client in Zig that outperforms official SDKs in its tests.