In launching adb (Arize database) we wanted to benchmark adb both internally as a database and at the system level in our application. Our goal is to show both the performance of the database that powers the application and the end application delivered user experience.
The benchmark tests cover these areas:
- Dataset Upload Programmatic
- Dataset Upload UI
- Trace Upload Programmatic
- Real Time Ingest to Read Time
- Search Over Large Data
The end application was tested under the following scenarios and we expect similar experiences under similar conditions:
- Internet connection speed test: 150MB Down / 150MB Upload
- MAC
Dataset Upload
The dataset upload tests how fast adb can upload and make available a large batch of data.
Dataset upload
The above shows a diagram of the dataset upload, a CSV or dataframe is uploaded into adb. The test shows how fast a large file can be uploaded into adb. The file upload uses a batch insertion datapath in adb that is designed for extremely fast ingestion of large files.
Trace Ingestion
The trace ingestion tests show how quickly adb can make individual realtime events available in the UI.
Trace Insertion
The trace insertion path utilizes the realtime ingestion path of adb going from event to user interface with subsecond timing.
Full Text Search
Full text search tests the ability to run search across a large number of spans with large chat text input and output attributes.
Full Text Search
Spans are ingested in adb with 25KB chat text strings in each span across two tests of 5M and 10M spans each. Full text search shows the search timing for regexp search across the data.
