napkin-math VS fast-sqlite3-inserts

Most production storage systems/databases built on top of S3 spend a significant amount of effort building an SSD/memory caching tier to make them performant enough for production (e.g. on top of RocksDB). But it's not easy to keep it in sync with blob...

Even with the cache, the cold query latency lower-bound to S3 is subject to ~50ms roundtrips [0]. To build a performant system, you have to tightly control roundtrips. S3 Express changes that equation dramatically, as S3 Express approaches HDD random read speeds (single-digit ms), so we can build production systems that don't need an SSD cache—just the zero-copy, deserialized in-memory cache.

Many systems will probably continue to have an SSD cache (~100 us random reads), but now MVPs can be built without it, and cold query latency goes down dramatically. That's a big deal

We're currently building a vector database on top of object storage, so this is extremely timely for us... I hope GCS ships this ASAP. [1]

[0]: https://github.com/sirupsen/napkin-math

Trying to estimate performance using some napkin math based on this: https://github.com/sirupsen/napkin-math

So napkin maths time. Typical cross-world bog-standard network speeds for a single TCP channel of ~25MiBps. A single HEADERS+RST pair is likely < 128 bytes (40 for the HEADERS + whatever payload, and 32 for the RST). So 8 pairs per K, 8K pairs per MiB, 200K pairs per 25MiB...

Yes, sequential I/O bandwidth is closing the gap to memory. [1] The I/O pattern to watch out for, and the biggest reason why e.g. databases do careful caching to memory, is that _random_ I/O is still dreadfully slow. I/O bandwidth is brilliant, but latency is still disappointing compared to memory.

[1]: https://github.com/sirupsen/napkin-math

https://github.com/sirupsen/napkin-math (memorize these)

I am experimenting with SQLite, where I try inserting 1B rows in under a minute. The current best is inserting 100M rows at 23s. I cut many corners to get performance, but the tweaks might suit your workload.

I have explained my rationale and approach here - https://avi.im/blag/2021/fast-sqlite-inserts/

the repo link - https://github.com/avinassh/fast-sqlite3-inserts

I am working on a project [0] to generate 1 billion rows in SQLite under a minute and inserted 100M rows inserts in 33 seconds. First, I generate the rows and insert them in an in-memory database, then flush them to the disk at the end. To flush it to disk it takes only 2 seconds, so 99% of the time is being spent generating and adding rows to the in-memory B Tree.

For Python optimisation, have you tried PyPy? I ran my same code (zero changes) using PyPy, and I got 3.5x better speed.

I published my findings here [1].

[0] - https://github.com/avinassh/fast-sqlite3-inserts

[1] - https://avi.im/blag/2021/fast-sqlite-inserts/

I have a side project where I tried to insert one billion rows in SQLite. I was able to insert 100 million rows using Python under 210 seconds. The same thing with PyPy took 120 seconds. I am wondering what kind of speed boost I would get with Cython

Measure, measure, measure! There is a PR which made really minor changes, but it got 2x speed boost with CPython version

[somewhat related] Recently, I was benchmarking SQLite inserts and I managed to insert 3.3M records per second (100M in 33 ish seconds) on my local machine - https://github.com/avinassh/fast-sqlite3-inserts Ofcourse the comparison is not apples to apples, but sharing here if anyone finds it interesting