<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>Vance Lin — Systems &amp; Storage Notes</title>
  <link href="https://vlx.this52.cn/feed.xml" rel="self"/>
  <link href="https://vlx.this52.cn/"/>
  <updated>2026-11-04T00:00:00Z</updated>
  <id>https://vlx.this52.cn/</id>
  <author><name>Vance Lin</name></author>
  <entry>
    <title>Batching WAL Fsyncs with io_uring Registered Buffers</title>
    <link href="https://vlx.this52.cn/posts/io-uring-batching-postgres-wal"/>
    <id>https://vlx.this52.cn/posts/io-uring-batching-postgres-wal</id>
    <updated>2026-11-04T00:00:00Z</updated>
    <summary>How grouping write-ahead log commits into fixed submission queues cut NVMe P99 commit latency by 41% under high-concurrency OLTP loads.</summary>
  </entry>
  <entry>
    <title>Tracing P99.9 Allocator Stalls in Go Services with eBPF Ring Buffers</title>
    <link href="https://vlx.this52.cn/posts/ebpf-ringbuf-tail-latency"/>
    <id>https://vlx.this52.cn/posts/ebpf-ringbuf-tail-latency</id>
    <updated>2026-10-19T00:00:00Z</updated>
    <summary>Using uprobe and sched_switch tracepoints inside a shared eBPF ring buffer to catch sub-millisecond stop-the-world stalls without profiler bias.</summary>
  </entry>
  <entry>
    <title>Bounded Clock Drift and Linearizable Lease Reads in Multi-AZ Raft</title>
    <link href="https://vlx.this52.cn/posts/raft-lease-reads-clock-drift"/>
    <id>https://vlx.this52.cn/posts/raft-lease-reads-clock-drift</id>
    <updated>2026-09-28T00:00:00Z</updated>
    <summary>Avoiding quorum ReadIndex round-trips on local reads by combining monotonic clock bounds with heartbeat lease invariants.</summary>
  </entry>
  <entry>
    <title>Training Zstandard Dictionaries for Sub-2KB Telemetry Events</title>
    <link href="https://vlx.this52.cn/posts/zstd-dictionary-small-payloads"/>
    <id>https://vlx.this52.cn/posts/zstd-dictionary-small-payloads</id>
    <updated>2026-09-11T00:00:00Z</updated>
    <summary>Standard compression struggles on tiny 400-byte JSON records. Pre-trained Zstd dictionaries lift compression ratios from 1.3x to 4.1x.</summary>
  </entry>
  <entry>
    <title>Diagnosing SQLite WAL Checkpoint Starvation Under Concurrent Readers</title>
    <link href="https://vlx.this52.cn/posts/sqlite-wal-checkpoint-starvation"/>
    <id>https://vlx.this52.cn/posts/sqlite-wal-checkpoint-starvation</id>
    <updated>2026-08-22T00:00:00Z</updated>
    <summary>Why a single long-running read transaction prevents WAL truncation, and how bounded read snapshots keep disk usage flat.</summary>
  </entry>
</feed>
