Login
You're viewing the techdon.dev public feed.
  • Sep 26, 2026, 5:12 PM

    back-bdb switched from the generic dbm-style API to the Berkeley-specific transactional API, which brought multithreaded concurrency support, fine grained locking, and ACID transactions. All of which brought along a lot of overhead too, and during development back-bdb was initially much slower than back-ldbm. We papered over that with a bunch of slapd-specific caching, to get it up to a usable speed. openldap.org/lists/openldap-de

    💬 1🔄 0⭐ 0

Replies

  • Sep 26, 2026, 5:15 PM

    Even though SleepyCat went away in 2006, we didn't start moving away from BDB until 2011, when we developed LMDB for ourselves. We probably would have stayed on BDB even longer, if Oracle hadn't started getting predatory about licensing. All in all I'm glad they did, glad they gave us the motivation to leave them behind.

    We didn't completely drop BDB until OpenLDAP 2.5, in 2021. It had a long run.

    💬 1🔄 1⭐ 3
  • Sep 26, 2026, 6:09 PM

    When I started hacking on OpenLDAP back in 1998 I was not a DB guy, I was a filesystems and network protocols guy. That's why my first contribution to OpenLDAP was back-ldap - my interest was in the network side of things. I got tangled up in back-bdb because I'm also interested in pure performance.

    Working with BDB taught me about DBs and transactions. I knew nothing about them before, had no interest. LMDB is a product of my filesystem and 8bit computing experience, married to BDB lessons.

    💬 1🔄 0⭐ 1
  • Sep 26, 2026, 6:13 PM

    Knowing about filesystems and virtual memory managers prompted my use of mmap. Knowing about the unpredictability of mmap flushing prompted my use of copy-on-write. Knowing about double-buffered graphics on the Apple ][ prompted my use of shadow paging (a common term in DB tech that I didn't learn until long after LMDB was released). Working with BDB taught me the value of ACID transactions - looking back, it's scary to think of a DB not having them.

    💬 0🔄 1⭐ 2