Databases. See also @AlexMillerDB@fosstodon.org or @alexmillerdb.bsky.app
- Tweets 1,042
- Following 211
- Followers 2,940
- Likes 409
I am here for the semantics.
Atomic commitment requires all nodes to agree. Consensus requires all non-faulty nodes to come to an agreement. If you have no faulty nodes, then atomic commitment can be consensus. Specialized, yes, but generally a useful one.
3
3
68
You can't always rely on O_DIRECT support existing. Docker-for-mac's FUSE FS thing doesn't have it. ZFS only *very* recently gained O_DIRECT support. If you're aiming for portability, it's wise to design for having to include explicit fsync calls.
1
2
38
arxiv.org/abs/2505.06501 has a couple diagrams that I think capture well their good filtered vector search taxonomy
1
15
1,877
[arXiv] Aurora DSQL: Scalable, Multi-Region OLTP
arxiv.org/abs/2607.13276
It's a DSQL paper, hot off the presses! Query processing in firecracker! Erasure coding on logs! The origin story of the turmoil crate! All cool stuff!
Can't sneak an arXiv upload past me, Marc. :p
3
23
130
9,058
Long time followers of brooker.co.za/blog/ are in for a fun game of connecting the dots between old blog posts and the DSQL paper contents. Is that erasure coding diagram looking familiar? It appeared 3 years ago on brooker.co.za/blog/2023/01/0… and now we get full context!
10
693
[VLDB '24] X-Stor: A Cloud-native NoSQL Database Service with Multi-model Support
vldb.org/pvldb/vol17/p4025-l…
Aside from the "everything can be implemented on top of a wide column store", there's a surprisingly detailed discussion of implementing modern multitenancy.
1
14
1,261
[ATC'22] DepFast: Orchestrating Code of Quorum Systems
usenix.org/conference/atc22/…
Quorum objects make for clean code. I really appreciate these "how to write distsys code in more correct and manageable ways" papers.
1
13
1,012
[arXiv] Beyond Chunk-Then-Embed
arxiv.org/abs/2602.16974
I had assumed that, when chunking (large) documents for embedding, adding context about the document to the chunk would always be superior, and it was interesting to see that's not *always* the case.
7
634
[arXiv] Beyond Cosine Similarity
arxiv.org/abs/2602.05266
I almost always see cosine, and it was interesting to read about similarity metrics as classes of more and less restrictive similarity measures.
4
52
413
31,011
South Bay Systems returns for its June meetup on June 26th! We'll have two speakers: @thinkingfish will talk about her experiences from working on Caching at Twitter and developing Pelikan, and Eric Liang will talk about Databrick's history-based data clustering feature.
1
1
7
1,554
The recording from this event is now available!
piped.video/9LiSWbRASKc
Our next South Bay Systems meetup is on May 26! This time, we're covering one of my favorite topics: databases, and how to use them to make better architectural decisions and build reliable systems.
We have two great talks lined up:
- "Building a Distributed Persistent Queue on FoundationDB": @HimankChaudhary will walk through how the queuing infrastructure at @TigrisData was designed and implemented.
- "Decisions, Principles, and Lessons from a Year of Teaching MySQL New Tricks": Steve Schirripa will share lessons and challenges around extending relational database systems at @VillageSQL.
Food and drinks will be provided courtesy of our hosts at @PingCAP. Registration link below.
9
914
Our next South Bay Systems meetup is on May 26! This time, we're covering one of my favorite topics: databases, and how to use them to make better architectural decisions and build reliable systems.
We have two great talks lined up:
- "Building a Distributed Persistent Queue on FoundationDB": @HimankChaudhary will walk through how the queuing infrastructure at @TigrisData was designed and implemented.
- "Decisions, Principles, and Lessons from a Year of Teaching MySQL New Tricks": Steve Schirripa will share lessons and challenges around extending relational database systems at @VillageSQL.
Food and drinks will be provided courtesy of our hosts at @PingCAP. Registration link below.
1
8
20
3,446
And the recording from this talk is now available!
piped.video/BG1zqnTGpyc
Thanks again to @cliff_click for the great talk, and LinkedIn for the great venue!
South Bay Systems returns for its April meetup on the 30th. This time we have @cliff_click giving a walkthrough of his teaching language for Sea of Nodes!
Sign up now! luma.com/nnq9aq27
2
26
5,606
South Bay Systems returns for its April meetup on the 30th. This time we have @cliff_click giving a walkthrough of his teaching language for Sea of Nodes!
Sign up now! luma.com/nnq9aq27
3
16
7,579
If you're an RSS user and a South Bay Systems attendee, I've added an RSS feed for the events at southbaysystems.xyz/atom.xml
4
588
Our next South Bay Systems meetup will be on March 31. We've got two awesome deep-dive talks:
- @ssougou will present deconstructing consensus and its application to Multigres
- @stuhood will talk about how modern full-text search is evolving toward columnar systems (and the tricky optimization challenges)
Food and beverages will be provided, courtesy of our hosts, @Snowflake
Come hang out with the systems crowd 👇
1
6
20
4,234
[CIDR '25] Linear Elastic Caching via Ski Rental
vldb.org/cidrdb/papers/2025/…
You should consider that holding a page in cache costs you, because RAM itself is expensive, and existing page replacement algorithms look at sizing cache independently (via miss-ratio curves).
2
1
20
1,284
There's a few papers which argue that DBMSs do page eviction wrong, and they always feel like incredibly compelling arguments. As a bonus, "Writeback-Aware Caching" pdl.cmu.edu/PDL-FTP/Storage/… points out that evicting a dirty page is more expensive than evicting a clean page.
2
298
[arXiv] Dynamic read & write optimization with TurtleKV
arxiv.org/pdf/2509.10714
TurtleKV shows a way to elastically move around the RUM conjecture space depending on what is important at the moment.
1
15
82
3,917
I continue to have a weakness for papers that show B-Trees and LSM-Trees as just two opposite tradeoffs along a continuum.
I have a personal fondness for papers/posts which present two very distinct and opposing designs as just two extremes of some spectrum of design trade-offs. LSMs vs B-Trees is a space in which I've seen a few rather different pitches of what that design spectrum could look like
5
659