Releases, internals, and benchmarks. Roughly one post a month.
The big one: rewrote the manifest engine, shaved 60% off steady-state diff time on a 100k-file pool. Also: new wire format, on-disk migration, breaking flag changes.
Four nodes, three different ISPs, one of them on flaky LTE backup. Throughput, retransmits, manifest churn, and where we still lose to plain rsync.
The old engine used a coarse-grained mutex around the in-memory manifest. Three readers and two writers were enough to make it the bottleneck. Here's what I replaced it with — and what almost worked but didn't.
RSS feed at /feed.xml. Or just check back occasionally — new posts go up first thing on the announcements list.
/feed.xml