Hacker News new | ask | show | jobs
by hit9 3791 days ago
Hi, skyline is a great work, here is my views:

1. Skyline has many built-in detection algorithms. Banshee only uses the 3-sigma rule.

2. Skyline's algorithms are pure and simple, without considerations of burr points, scatters (like counters).

3. Banshee comes with alerting rules management panel. Skyline's webapp is a simple analyzation results viewer.

4. Banshee uses the anomaly-factor trending (via weighted moving average) and skyline doesn't. The trending follows the analyzed score (which is to describe how anomalous the current metric is), thus, alerts only happen if the metric is anomalous enough (i.e. with a very big value) or if anomalous time is long enough.

5. Banshee has the management for projects, users, rules.

6. Banshee uses statsd as data source, skyline uses carbon.

7. Skyline uses redis as storage. Banshee has no external storage server dependencies. Metrics are stored on disk via leveldb, and rules are stored on disk via sqlite.

8. Banshee is young and skyline is no longer actively maintained...

1 comments

Thanks for the break-down. It looks interesting. At work, we have been testing Skyline for a number of different metrics coming in from a large distributed network. I'm interested in seeing how Banshee performs in there.