Benchmark

Benchmark are set to quicly evolve. I have currently multiple ideas to improve performance.

Command

You can run zig build benchmark if you clone the repo to benchmark your machine. More info.

Here an example on my machine:

=====================================

Populating with 500000 users.
Populate duration: 8.605314 seconds

Database path: benchmark
Total size: 50.99Mb
LOG: 0.00Mb
BACKUP: 0.00Mb
DATA: 50.93Mb
  User: 50.92683124542236Mb 500000 entities

--------------------------------------

Query:          GRAB User {}
Duration:       457.758686 ms

Query:          GRAB User [1] {}
Duration:       1.285849 ms

Query:          GRAB User [name] {}
Duration:       138.041888 ms

Query:          GRAB User {name = 'Charlie'}
Duration:       63.094060 ms

Query:          GRAB User {age > 30}
Duration:       335.654647 ms

Query:          GRAB User {bday > 2000/01/01}
Duration:       52.896498 ms

Query:          GRAB User {age > 30 AND name = 'Charlie' AND bday > 2000/01/01}
Duration:       56.295173 ms

Query:          GRAB User {best_friend IN {name = 'Charlie'}}
Duration:       69.165272 ms

Query:          DELETE User {}
Duration:       93.530622 ms

=====================================

Cloud

I started to benchmark using fly.io

Using a 1x shared cpu I get around 40ms to parse and write 5000 users, not bad.

File Parsing

The majority of time taken by query come more from writting the response rather than parsing files.

In this example I create a random dataset of Users using this schema:

User (
  name: str,
  age: int,
  email: str,
  bday: date,
  last_order: datetime,
  a_time: time,
  scores: []int,
  friends: []str,
)

Here a user example:

run "ADD User (name = 'Diana Lopez',age = 2,email = 'allisonwilliams@example.org',scores=[37 85 90 71 88 85 68],friends = [],bday=1973/11/13,last_order=1979/07/18-15:05:26.590261,a_time=03:04:06.862213)

Let's do a query that parse all file but dont return anything, so we have the time to read and evaluate file but not writting and sending output.

run "GRAB User {name = 'asdfqwer'}"

50 000 000 Users

This take 6.24GB space on disk.

Thread	Time (s)	Usage (%)
1	40	10
2	21	18
3	15	25
4	12	30
6	8.3	45
8	6.6	55
12	5.1	85
16	4.3	100

1 000 000

This take 127MB space on disk, sperated into 24 files of 5.2MB

Thread	Time (ms)
1	790
2	446
3	326
4	255
6	195
8	155
12	136
16	116

2.7 KiB Raw Blame History