Summary of experiments
Executive summary
Decision Matrix
| Criteria | Deepclone | XFS | BTRFS | OverlayFS |
|---|---|---|---|---|
| Can the solution be rolled out to all platforms ? | Yes | |||
| Does the solution have any limitations ? Please list the required limits (e.g. maximum N snapshots) | Yes. Approximately 1500 git files and git directories, above which snapshot performance is above latency target | |||
| Does the solution require a migration ? | No | |||
| Does the solution scale ? Please list any scaling limitations (e.g. resource usage rises linearly with number of transactions, or repository size) | No. Snapshot performance scales linearly (or worse) with .git file count and git directory count | |||
| Does there need to be a mode where transactions are disabled ? | Yes, due to snapshot performance issues | |||
| Does the solution meet latency targets ? | No | |||
| Is the solution performance with reftables enabled ? | No |
For the above approaches, we are running various experiments for.
Latency targets
- 500ms for main methods. 1 second is tolerable.
- 10 seconds for some slow methods . 30 seconds is tolerable.
Experiment links
Add a link to the experiment report for further detail.
- Experiment name:
<link> - Experiment name:
<link>
Edited by Thong Kuah