TigerBeetle Core System Architecture: Deconstructing Performance Engineering

This was always the plan, and if you look closer at VSR and the state machine interface you’ll see it’s already pluggable. We just haven’t packaged it. (We’re dogfooding our first few internal “CustomBeetles” before we package and document.)

Emphasis on: opinionated custom databases. One database might be SQL-based for periodic report generation. Another database might be a noSQL key-val store designed to effortless grow with the number of end-users. General-purpose programming languages already cater to the 'own business logic' part - it's their whole job. What's left? System architecture, networking features, consensus. That's Kafka.

llvm for databases?

I'm pretty sure they manage to get that level of performance and reliability, since they have a very limited schema.

I'm pretty sure you can't just do a precise 128 byte align, if one of the element is an image blob or varchar(1000).

Hey, very inspiring article, Redis engineer here. How do you work with static allocation on variable query structure, and row counts that can explode depending on the data shape?

And isn't there a benefit for small allocations on advanced memory allocations that you can't leverage if all is working in big page allocations? Do you implement memory allocations from scratch or leveraging existing allocator on top of these memory blocks strategy somehow?

If one has a TigerBeetle cluster with high inter node latency, are there any easy wins to lower the latency of the whole cluster left? My head hurts when I think of latency in large clusters, so your work this year with latency was inspiring.

EDIT: I guess part of the question is about the problems with clusters with >130ms latency and if there are challenges you consider easy.

Hey Joran, awesome stuff. I see a TB post every now and then and it seems like such an interesting problem space to work in. I'm all the way at the other end of the stack, most software we write day-to-day is in JVM-based languages where you don't have to think about these things at all (we get by with ms latencies instead of ns). Reading this post inspires me explore low-level engineering more and I figure zig or rust would be a good place to start.

What are some interesting problems or things you can think of to work on that would give someone new a nice amount of exposure to this kind of programming?

How has AI changed the way that TigerBeetle does software engineering? Given the project’s idiosyncratic language/memory allocation choices, it’s an interesting data point how well the frontier models work for you guys.

Great and interesting work !

What use cases is Tiger Beetles architecture not a good fit for?

When tiger beetle becomes pluggable to different use cases, how should I think about "do I want tiger beetle?"

Tobi here from TB. Great question, and it's important to be nuanced here.

It really depends on the problem space. For example, many OLAP workloads (analytical) contain large amounts of parallelizable work, then multi-core execution is absolutely the way to go. That also aligns well with the direction CPU technology is taking, with core counts continuing to increase.

For the transactional workloads we see at TigerBeetle, and in other transactional systems I've worked with, the picture is quite different. We see a lot of read-modify-write operations combined with a power-law distribution of the data.

Take a simple banking example: some accounts, such as those belonging to large online retailers, see much more activity than the average individual account. You might have 80 - 90% of transfers touching a relatively small number of these hot accounts.

Operations on the same account must be serialized to preserve correctness. That means this part of the workload cannot be meaningfully parallelized. In fact, attempting to parallelize it can make performance worse because of lock contention and coordination overhead, something the "Universal Scalability Law" captures quite well (but is also easy to test out yourself with a simple experiment).

Instead, we focus on batched execution. We carefully structure execution to make effective use of CPU caches and efficient algorithms, so that a single batch can be processed extremely efficiently without any coordination. Batch execution also allows to amortize I/O and replication.

That being said, there are areas where we could use multi-threading (e.g. compaction) that are not on the hot execution path.

Ha, I was explaining this just yesterday to a few people at $WORK (I'm not at TB, though we do use a bunch of Zig in prod).

"Multiprocessing" (multiple CPU cores independently executing and only able to coordinate via some message-passing system -- a definition which encompasses both multi-core CPUs and horizontally scaled distributed systems, contrasting slightly with the normal definition) is challenging for a few reasons.

Firstly, the details are an open math question, but I'll blindly state that some problems aren't amenable to parallelization. I.e., no algorithm can meaningfully improve performance via parallelization no matter the implementation. Think through how you would more quickly compute hash(hash(hash(hash(...)))) for example. The serial dependency makes things challenging. That isn't too dissimilar to the problem TB faces.

Secondly, message passing is expensive. If the only way two CPU cores can coordinate is through a multi-level cache, at best you're incurring ~tens of nanoseconds of latency per message. Contrast that with a base rate of 512 bytes processed per nanosecond with enough attention to detail on typical modern server hardware (4 pipelined AVX512 instructions at 2GHz). Messages are several orders of magnitude worse than your normal work, so if you need very many of them then you're hosed from a performance perspective (worse with longer delays, like networked computers). Even very parallelizable problems at an abstract level can suffer performance losses by trying to add even one extra core. This blog post [0] doesn't perfectly capture the idea, but it's close (and a fun read regardless).

Thirdly, message passing is an insanely complicated programming abstraction to reason about. My first two points were more about what TB was saying -- realities of modern hardware -- but the programmer experience is important too (even if you don't believe that post-2020, the LLM experience doesn't differ much from the human experience; bad code begets more bad code, slowly). The core mechanism for correctness in most software is being able to reason about "this thing is true, therefore that thing is true" and iterating. You rely on invariants like "this is sorted" to build other working theories. The invariants in multiprocessing code are much more nebulous and less amenable to accidental discovery, also less amenable to being able to build or compose them into other stronger invariants as you add code. The main reason for that is that you know almost nothing about the relative order of those messages with respect to each processer's view of which instructions happened when (and for purely multi-core "multiprocessing" the story is even worse; while my description of message-passing being the core primitive is correct, that's not what's exposed to you as a programmer, and different memory models can have even weirder interleavings than your code would naively suggest -- i.e., your code is being decomposed into smaller subunits than even a single assembly instruction, and the message passing happens at that level). The combinatorial explosion (an exponential explosion really, but big numbers either way) of states you might be interacting with makes it very difficult to understand _anything_ about the system you're examining. That's why you see a handful of primitives used over and over -- if you can decompose your problem into a parallel map plus an associative reduce then you can probably figure out some way to make it better through parallelization (not always, especially if the framework is too generic, see the linked blog post [0] if you weren't enticed to read it previously). If you can't decompose it into know primitives then it's an open research problem every time.

The crux of that third point (and we could definitely add more explanation and additional problems) is that there's a huge cost to multiprocessing. You have to be buying something substantial to even want to reach for it, else you have to be in one of the "easy" problem spaces where somebody else has done the hard work (e.g., stateless webservers).

TB isn't that. Their whole raison d'être is state management, and not in a way that's easily amenable to parallelization.

[0] https://adamdrake.com/command-line-tools-can-be-235x-faster-...

If you give the TigerBeetle client a single transfer, it sends it off immediately to the cluster. There's no delay. No Nagle!

But if your application then creates another transfer against the client, and another, while the first request is inflight, then the client will autobatch under the hood and send these off as a batch when the first request returns.

You get this sweetspot then between latency and throughput. And your latency is not spiking as your load increases, since your throughput is now able to keep up.

I think you design in layers, frontends that work as clients to TigerBeetle for work in batches (as mentioned by the sibling comment), but the whole idea of removing latency differences by removing unpredictability means that you don't get the jitter of latency differences that can cause backing up in normal scenarios.