tigerbeetle VS sb_lower_bound

I'm waiting for someone to implement the Redis API by swapping out the state machine in TigerBeetle (which was built modularly such that the state machine can be swapped out).

https://tigerbeetle.com/

I fully agree with what Prime says at the end - Joran has really set a new bar here for all future database presentations.

Hearing that the entire TigerBeetle domain logic lives in a single file [0] (and is intended to be pluggable for other OLTP use cases!) makes it 1000% more tempting to spend the weekend getting up to speed with Zig.

[0] https://github.com/tigerbeetle/tigerbeetle/blob/main/src/sta...

Why would you want to build your own accounting ledger from scratch? Accounting is a completely new domain for most engineers, and TigerBeetle (https://tigerbeetle.com/) already solves this problem.

Maelstrom [1], a workbench for learning distributed systems from the creator of Jepsen, includes a simple (model-checked) implementation of Raft and an excellent tutorial on implementing it.

Raft is a simple algorithm, but as others have noted, the original paper includes many correctness details often brushed over in toy implementations. Furthermore, the fallibility of real-world hardware (handling memory/disk corruption and grey failures), the requirements of real-world systems with tight latency SLAs, and a need for things like flexible quorum/dynamic cluster membership make implementing it for production a long and daunting task. The commit history of etcd and hashicorp/raft, likely the two most battle-tested open source implementations of raft that still surface correctness bugs on the regular tell you all you need to know.

The tigerbeetle team talks in detail about the real-world aspects of distributed systems on imperfect hardware/non-abstracted system models, and why they chose viewstamp replication, which predates Paxos but looks more like Raft.

[1]: https://github.com/jepsen-io/maelstrom/

[2]: https://github.com/tigerbeetle/tigerbeetle/blob/main/docs/DE...

> There is no reason to use a memory unsafe language anymore, except legacy codebases, and that is also slowly but surely diminishing. I'm still yet to hear this amazingly compelling reason that you just need memory unsafe languages. In terms of cost/benefits analysis, memory unsafety is literally all costs.

Tell that to the authors of new memory unsafe languages (like Zig) and creators of new project in those languages (like https://tigerbeetle.com) :(

Then you'll want to look at https://mhdm.dev/posts/sb_lower_bound/#prefetching

100mb is large enough that the branchy version turns out to have a slight advantage, more due to quirks of x86 (speculative execution) rather than being better.

"very similar topic" is an understatement. Funnily enough the "implementation to perform the best on Apple M1 after all micro-optimizations are applied" in the Conclusion is equivalent in terms of the how many actual comparisons are made as with sb_lower_bound. Out of curiosity I've benchmarked the two and orlp lower_bound seems to perform slightly worse: ~39ns average (using gcc) vs ~33ns average of sb_lower_bound (using clang -cmov). I'm comparing best runs for both, usual disclaimer of tested on my machine.