I've seen the usual marketing fluff about "lightning-fast queries" from both camps, but the actual numbers always seem to vanish into a cloud of asterisks and "it depends." The benchmarks they publish are inevitably on pristine hardware, with perfectly optimized schemas, and queries that look nothing like my production workloads.
Has anyone done a real, apples-to-apples comparison recently? I'm particularly interested in:
* **Time-range sensitivity:** Does one degrade more gracefully than the other when querying 30 days vs. 3 hours of high-cardinality data?
* **Concurrency under load:** Their docs talk a big game about performance, but what happens with 20+ concurrent dashboards refreshing? I've been burned before by systems that crumble the moment you look at them sideways.
* **The hidden tax of aggregates:** Everyone shows off a simple `MAX()` or `MEAN()`. Try throwing a `PERCENTILE()` or a complex derivative into the mix and watch the latency charts look very different.
I'm skeptical of any vendor-provided TPC-like benchmarks. I want to know what the trade-offs are when the rubber meets the road on a standard cloud instance, with real-world cardinality and mixed query patterns. What's the actual performance delta once you account for their respective compression, indexing strategies, and the inevitable memory overhead?
Vendor claims are hypotheses, not facts.