> ## Documentation Index
> Fetch the complete documentation index at: https://arize-ax.mintlify.site/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# adb performance & benchmarks

> Benchmark results for adb across dataset uploads, trace ingestion, and full-text search, compared to other observability platforms.

adb is built for petabyte-scale performance with sub-second latency. The benchmarks measure the parts of the system users feel most directly: how fast data uploads, how quickly a trace goes from ingested to visible, and how search holds up over large datasets.

## What was measured

The benchmark suite covers five categories:

1. **Dataset upload (programmatic):** batch file ingestion via code.
2. **Dataset upload (UI):** the same upload through the user interface.
3. **Trace upload (programmatic):** ingesting individual trace events.
4. **Real-time ingest-to-read:** latency from an event arriving to it being visible in the UI.
5. **Search over large data:** full-text search across very large span datasets.

## Methodology

Tests were run over a 150 MB/s down / 150 MB/s up connection. Full-text search was measured as regexp search over datasets of 5M and 10M spans, each span carrying a \~25 KB chat-text string.

The `adb (P50)` column is the median database timing. `Arize AX Platform` is the end-to-end experience delivered to the user (database plus application). `Platform A`, `Platform B`, and `Platform C` are other observability platforms run through the same tests.

## Results

| Benchmark                          | Rows       | adb (P50) | Arize AX Platform | Platform A            | Platform B                                             | Platform C                                                               |
| :--------------------------------- | :--------- | :-------- | :---------------- | :-------------------- | :----------------------------------------------------- | :----------------------------------------------------------------------- |
| Dataset CSV upload (code)          | 50k        | 166ms     | 11.95s            | *Rate limit exceeded* | 312.21s (26× slower)                                   | 76.03s (6.3× slower)                                                     |
| Dataset CSV upload (code)          | 200k       | 545ms     | 13.93s            | *Rate limit exceeded* | *Not supported: dropped 149,999 elements (full queue)* | *Not supported: content length 147,270,760 B exceeds 20,971,520 B limit* |
| Dataset CSV upload (code)          | 500k       | 612ms     | 26.85s            | N/A                   | *Not supported*                                        | *Not supported*                                                          |
| Dataset CSV upload (code)          | 1M         | 681ms     | 36.54s            | N/A                   | *Not supported*                                        | *Not supported*                                                          |
| CSV upload (UI)                    | 50k        | 148ms     | 16.24s            | *Max file size 10 MB* | *Crashed after 2 hrs (\~24k rows)*                     | 65.52s (4× slower)                                                       |
| Spans ingest (ingestion + UI load) | 1k traces  | 15.57ms   | 0.13s             | 0.36s (2.7× slower)   | 23.48s (176× slower)                                   | 0.26s (2× slower)                                                        |
| Spans volume test                  | 50k traces | 15.42ms   | 4.45s             | 17.48s (4× slower)    | 1,238.33s (70× slower)                                 | N/A                                                                      |

<Callout type="info">
  Cells marked *Not supported*, *Rate limit exceeded*, or *Crashed* reflect limits the competing platform hit at that scale during testing. *N/A* means the case wasn't run for that platform.
</Callout>

## What the numbers show

Across every category, adb's raw database timings stay in the sub-second range even as row counts grow into the millions, and the end-to-end Arize AX platform stays fast where other platforms rate-limit, cap file sizes, drop records, or fail outright. The gap widens with scale: several platforms that keep pace on small trace batches become tens to hundreds of times slower, or stop working, as data volume climbs.

adb keeps your traces, evaluations, and annotations in open Iceberg format, so [Data Fabric](/ax/security-and-settings/data-fabric) can continuously sync that data straight to your own cloud warehouse, like BigQuery or Snowflake, for analytics and custom workflows.

<Card title="Sync your data with Data Fabric" icon="arrow-right" href="/ax/security-and-settings/data-fabric" />