We do not stop at reading papers. We measure how storage really behaves, reimplement the ideas we study, and publish everything openly so others can check our work.
00
Published
01
In progress
00
Open proposals
04
Research areas
We compare io_uring's completion-based model with epoll-style readiness notification for random-read heavy workloads, measuring throughput, tail latency, and CPU cost per I/O across queue depths.
By I/O Working Group
Write-up coming soonOur research follows the path of a query, from the execution engine down to the disk.
Completion vs. readiness, io_uring, epoll, and user-space I/O stacks.
B+ trees, LSM-trees, page layouts, compaction, and write amplification.
Page caching, eviction policies, and when to bypass the OS page cache.
Volcano iterators, vectorization, and compiled query plans.
Published work, projects underway, and open questions looking for contributors.
Every study here started as a question from a member. No prior research experience needed, just curiosity and patience.
Start a proposalBring a question worth measuring. Open a proposal on GitHub or pitch it at a session.
Members join in, agree on a method, and build the benchmark or prototype together.
Results are peer reviewed by the community before anything is published.
We release the write-up, the data, and the code, so anyone can reproduce it.
