Howard Chu Profile picture
CTO Symas Corp., Chief Architect OpenLDAP Project, Musician
Sep 12, 2020 5 tweets 2 min read
Common wisdom was that it's impossible to design an ASIC-proof Proof of Work algorithm. tevador, SChernyk, and I proved that bit of "wisdom" wrong with RandomX. Folks who understand hardware design actually get it.
np.reddit.com/r/Buttcoin/com… We started over 2 years ago from the simple insight that ASICs can only optimize individual algorithms, and thus we needed a dynamic approach. old.reddit.com/r/Monero/comme…

Months of prototyping and testing went in before we arrived at the current design.
May 6, 2019 4 tweets 2 min read
@jpmens It's complicated™

Symas is still working on this. But the Samba LDB modules are unsuitable for use within slapd. We are, unfortunately, reimplementing most of them as slapd overlays, tho still reusing as many Samba libs as possible.

Samba team won't accept OpenLDAP code. @jpmens Understandably; they lack expertise to maintain it. Just as OpenLDAP Project would reject code we cannot maintain.
Oct 20, 2018 13 tweets 2 min read
Today's OpenLDAP runs on a handheld PDA over wifi faster than most other LDAP packages on big servers with dedicated wired networks. Designing software for efficiency is more important than designing for scale. Achieving efficiency gives you scale for free.
Oct 6, 2018 4 tweets 3 min read
@fanf42 @walterbio LMDB is significant of course, but even back-bdb/hdb had dropped latency down to 0.8ms. There were huge improvements all across the stack. Faster encode/decode in liblber, faster processing in libldap, complete restructuring of slapd internal APIs, etc. @fanf42 @walterbio The result of years of intensive profiling and testing.

Some simple rules throughout, of course: avoid memcpy, avoid malloc, avoid strlen. These 3 functions used to be 75% of execution time. E.g. liblber now does zero-copy decode.