What the study found
The study found that Replay-as-a-Service (RaaS), a technique for storage-disaggregated online transaction processing (OLTP) databases, reduced long tail latency. The authors report that it also improved overall throughput in their implementation and evaluation.
Why the authors say this matters
The authors say this matters because storage-disaggregated databases are widely used in the cloud for benefits such as better resource use, less fragmentation, separate scaling of compute and storage, and cost savings. They state that RaaS addresses a significant limitation they identified in these systems: long tail latency caused by the log-as-the-database design.
What the researchers tested
The researchers focused on OLTP storage-disaggregated databases such as Amazon Aurora, Microsoft Socrates, and Neon. They introduced RaaS, which decouples log replay logic from the storage engine and makes it an independent service that can use idle servers or dedicated servers in the cluster.
What worked and what didn't
In the authors' implementation in OpenAurora, an open-source storage-disaggregated database based on PostgreSQL, RaaS reduced P95 tail latency by 40.1% and increased overall throughput by 75.9% in SysBench experiments. The abstract does not report any approach that worked poorly or failed.
What to keep in mind
The available summary does not describe limitations, caveats, or negative results beyond the general latency problem being addressed. The reported results are from OpenAurora and SysBench experiments, so the abstract does not show whether the same gains would appear in other systems or workloads.
Key points
- RaaS is presented as a way to reduce long tail latency in storage-disaggregated OLTP databases.
- The authors say the latency problem comes from the log-as-the-database design and long log replay chains.
- RaaS decouples log replay from the storage engine and can use idle or dedicated servers in the cluster.
- In OpenAurora, RaaS reduced P95 tail latency by 40.1% in SysBench experiments.
- The abstract reports a 75.9% improvement in overall throughput.
Disclosure
- Research title:
- Replay-as-a-Service reduces tail latency in storage-disaggregated databases
- Publication date:
- 2026-04-02
- DOI:
- 10.1145/3786688
- OpenAlex record:
- View
Get the weekly research newsletter
Stay current with peer-reviewed research without reading academic papers — one filtered digest, every Friday.