Performance

The shape of a command.

Zinn runs when you call it, does its work and exits. These are the numbers for that round trip, and the path they take.

100 tasks · 1 comment each · CLI 0.44.0 · macOS arm64

Inspector

Flow

Read

$ zinn task list --json

One command from an agent to SQLite and back. The JSON is what the agent reads.

Drag nodes. Scroll to zoom. Double-click to frame.

Time split

120 ms

Node boot41 ms
Module load38 ms
CLI12 ms
Core2 ms
SQLite3 ms
Disk1 ms
Output1 ms
Exit22 ms

0.20 mWh above idle. 0.06 mWh of it Node.

Command and package figures are measured. Node timings and install timings are placeholders.

p50

0.12 s

Peak memory

82 MB

Output

71.5 KB

Database

471 KB

Commands

What a command costs.

The 100-task fixture, one comment per task. p50, p95, peak memory and output are measured for all seven commands.

Command	p50	p95	Peak memory	Output
zinn task list	117 ms	124 ms	81 MB	6.6 KB
zinn task list --json	117 ms	119 ms	82 MB	71.5 KB
zinn task view BNCH-50	113 ms	115 ms	79 MB	737 B
zinn task create	157 ms	161 ms	81 MB	66 B
zinn task move BNCH-50 TODO	156 ms	159 ms	81 MB	62 B
zinn task search amarillo BNCH	116 ms	117 ms	80 MB	390 B
zinn backup	155 ms	160 ms	80 MB	140 B

Project size

What a bigger board costs.

The same list command as the board grows. The 100, 1,000 and 10,000-task rows are measured.

Tasks	Database	List time	Peak memory	JSON output
100	471 KB	117 ms	81 MB	71.5 KB
1,000	3.1 MB	160 ms	93 MB	715.9 KB
10,000	29.0 MB	3.45 s	187 MB	7.2 MB

Storage

What the database holds.

Empty database: 123 KB
Per task: 1.6 KB
Per comment: 1.3 KB
At 10,000 tasks: 29.0 MB

An empty database is 123 KB. A task adds about 1.6 KB and a comment about 1.3 KB. At 10,000 tasks the file is 29.0 MB.

Method

What these figures are.

Measured with CLI 0.44.0 and the packages it installs. Command times are medians of 30 runs, p95 is the nearest rank of those 30, and peak memory is the median per-run peak RSS from hyperfine. Each task has a 40-character title, a 200-character description and one 200-character comment.

Node timings and install timings are placeholders. Estimated processor energy above idle, on an M4 MacBook Pro. The read flow uses about 0.20 mWh above idle, and Node accounts for 0.06 mWh.