24 karma · joined September 22, 2017
People did get the spirit of the question and have provided helpful responses, including context about their query structure/complexity.
While reviewing IoT databases I saw a lot of discussion here and elsewhere about database performance and wanted to see some numbers. Of course, applications with different workloads will have different database needs, but it is still helpful to hear that 5000 rps were served with ABC database with XYZ hardware assuming that the majority of queries on the majority of apps are simple lookups.
"512gb of ram and dual EPYC 74F3 24 core processors, backed by a 4 drive raid10 nvme linux software raid volume on top of micron 9300 drives"
roughly translates to about an db.r8g.16xlarge (64 vCPUs, 512gb ram) $4,949 / month on-demand for compute
I'm not familiar enough with hardware to determine IOPS for the raid config but I believe it is greater than the maximum for io2 block express storage on aws (256k IOPS):
$0.10 per provisioned IOPS-month = 256000$.10 = $25,600 / month IOPS -- which feels high so I might be way off on the raid setup's IOPS
$0.125 per GB-month storage = 500gb $0.125 = $62.50
That's about $31,930 / month without any reserved discounts for an estimated capacity of 5,000 rps, sound about right? Would you say your total hardware cost is less than one or two months of comparable compute on AWS if the above is true?
Separately, I know many startup founders who have long-term software developers ("contractors", apparently...) in other countries like India, Brazil, etc. Paying them with whatever random popular money transfer service. It's common practice but is it legal or risky on the U.S. side?