Comparison
The same networked cluster, the same data, the same Windows machine. Here are the results, how we got them, and the raw data to redo the maths.
12 s against 52 s for the runner-up and about 4-5 minutes for the others
Open the program, see the 10,003 topics, search for a value across 2 million messages.
Sum of the median startup, topic list and search. Kafbat UI did not finish the search.
4× faster than the runner-up, 20× or more than the others
A tax code across 2 million JSON messages, 972 MB: 200 expected matches.
Kafbat UI stopped after 5 minutes with 160,200 messages read. AKHQ and Redpanda Console: one run; AKHQ with a server setting changed to reach the end.
0.88 s to see all 10,003 topics
From the click to the list on screen.
Offset Explorer from the command line. AKHQ shows the first page. Kafbat UI is not charted: between 3 and 58 s, highly variable. KafkIO also counts the messages of every topic.
1.5 s from double click to a ready window
For web tools, the server startup counts.
AKHQ, Kafbat UI and Redpanda Console: server startup, one run.
1.5 s to read and show 10,000 messages
The last 10,000 of a 2-million-message topic.
Offset Explorer from the command line. Redpanda Console reads the first 10,000. AKHQ and Kafbat UI read in pages of 50 and 100.
469 MB at the search peak, window included
Here the one using less wins: Topicup is third.
Working set of all processes. AKHQ: server only, without the browser. Redpanda Console: browser included. Kafbat UI not measured. Topicup: the Rust engine uses 25-130 MB, the rest is WebView2.
33 questions with the expected result known in advance: compression codecs, Unicode, a 1 MB message, regex, key, headers, transactions, compaction, JsonPath.
Messages of an aborted transaction
Topicup is right
An aborted transaction never happened.
Topic with transactions, 40 committed
Topicup is right
Same reason: the 20 aborted ones do not count.
Regex with .* on a 1 MB message
Topicup is right
Rust's regex engine runs in linear time.
Search by key, hand-assigned partitions
Offset Explorer is right
Topicup reads only the computed partition. To be improved.
Not measured: Kadeck, which requires registration, and products whose licence does not allow publishing comparisons.
Topicup and the other program one after the other, in the same time window, so a slow moment on the network weighs on both.
Only one program at a time on the client. Web servers run on the same machine and read the cluster over the network.
From the command or click to what the program shows or reports.
Of 3 runs; where there is only one run we say so next to the number.
Across 12 million messages, 5.8 GB, the broker's disk becomes the limit: Topicup 1 min 39 s, Offset Explorer 1 min 50 s. About 10%.
Offset Explorer uses less memory: 145 MB at rest and 372 MB during the search, against 293 and 469 MB for Topicup, mostly the window (WebView2).
| Offset Explorer: search | Topicup (command line) | Offset Explorer (-findMessages) |
|---|---|---|
| Total time, from launch to exit | 9.6 s (9.3-18.3) | 40.2 s (39.8-41.4) |
| Search only, as the program reports it | 8.8-9.1 s (16.4 in the slow run) | 32.6-34.2 s |
| Messages searched and found | 2,000,000, 200 | 2,000,000, 200 |
Topicup's slow run, 18.3 s, had the network at 22.6 ms instead of 11, with the broker still busy writing other data.
| KafkIO: user interfaces | Topicup (app) | KafkIO |
|---|---|---|
| Startup, up to the connection list | 1.53 s (1.38-1.53) | 11.5 s (11.1-11.7) |
| Connect and list the 10,003 topics | 0.88 s (0.88-0.89) | 51.3 s (51.2-53.0) |
| 10,000 messages read and shown | 1.50 s (0.94-1.52) | 6.3 s (6.2-6.3) |
| Search across 2 million messages | 9.9 s (9.7-10.2), 200 of 200 | 201 s (193-201), 200 of 200 |
| RAM at rest | 293 MB | 275 MB |
| RAM after the topic list | 351 MB | 1,049 MB |
| Peak RAM during the search | 469 MB | 1,898 MB (up to 2,023) |
KafkIO fetches the topics with the message count of each (27,026 partitions); Topicup shows the names with one request and loads counts when you open a topic.
One JSON line per measurement, as the script wrote it. Only the test PC's paths were removed.
Offset Explorer, Kafka Tool, KafkIO, Kafka Magic, AKHQ, Redpanda Console and Kafbat UI are trademarks of their respective owners. The names are used only to say what we measured: Topicup is not affiliated with or endorsed by their makers.