Member OP
So I signed up with Contabo (you know the one) for their "dedicated CPU" plan six months ago (or was it seven? Time blurs when you're debugging at 3am) and everything seemed fine (if you call "fine" a baseline that wobbles like a jelly on a washing machine) until I started running actual benchmarks for a client project (the kind where you need reproducible numbers, not lottery results).
Here's the smoking gun. [code]Run 1: sysbench cpu 1834 events/sec
Run 2: sysbench cpu 947 events/sec
Run 3: sysbench cpu 2102 events/sec
Run 4: sysbench cpu 891 events/sec
That variance (if you can call a 2.3x spread "variance" and not "someone else's cryptocurrency operation next door") is not silicon lottery. That's neighbor noise. I opened a ticket (politely! I was polite! My mother raised me right) and after three weeks of "our technicians are investigating" they admitted (in writing, which I screenshot, because of course I did) that "dedicated" on their platform means "dedicated thread on a shared core" (their words, not mine (though I wish I'd coined something that brazen)).
I migrated to InterServer (true dedicated, verified with lscpu showing physical cores I can actually touch (metaphorically)) at roughly 40% premium. Here's the after: [code]Run 1: sysbench cpu 2847 events/sec
Run 2: sysbench cpu 2833 events/sec
Run 3: sysbench cpu 2851 events/sec
Run 4: sysbench cpu 2845 events/sec
Standard deviation under 0.4%. That's what dedicated means. That's what I paid for. Consider this a warning (a polite one (naturally)).
