This message was deleted.
# general
s
This message was deleted.
s
The first thing that pops out is the fact that the test queries do not use the __time filter. This means that all queries will process all segments, no pruning and there is no mention of secondary partitioning either. There is also no mention of the segment granularity used or whether the data in Druid was compacted. Seems to me that they've optimized the data model and queries for StarRocks but not for Druid.
šŸ™ 1
j
I've done RDBMS competitive performance benchmarking in one of my past lives at Oracle ... the people involved and tips/tricks used can affect the benchmark results as much as the product itself. I am pretty sure if they involved us in the benchmark and allowed us to configure and run the Druid tests, the results would look much different šŸ˜‰.
šŸ™ 1
m
Good answers, Sergio and John. I researched Kenan a bit just to make sure he didn't work for StarRocks... šŸ˜‰
šŸ˜‚ 1
Here's my recommendation: • If you're interested in StarRocks' claims, try to reach out to their community or support channels for additional insights and clarifications on the benchmark specifics. • Consider running your own comparative tests on smaller datasets using workloads relevant to your use case. This allows for a more tailored comparison under your specific conditions. • Don't rely solely on performance as the sole decision factor. Other aspects like feature set, ease of use, cost, and community support also play a role in choosing the right technology for your needs.
šŸ‘ 2