This message was deleted.

Hi Julian, If you're looking at infrastructure costs, then if you don't mind having a "simpler" UI front-ending Druid, you might want to also take a look at Imply Polaris for cost and performance comparison. Polaris has some smaller project sizes within the price ranges you are quoting, I am not sure how performance compares but you should be able to get a free trial and test it out. Thanks. John

We run druid ourself but it might just be over provisioned. Also, wonder what query latency is for AWS case, we expose an API to customers where query latency should be at around 100-200ms, that might be the case their infra costs less and the fact they use arm64 instances and we use amd64

By latency do you mean Router/Broker reported query time? Or the added latency between the client and the Broker/Router? If the former --that would be completely dependent on the workload, so unfortunately something you would just have to test. But as an example I've attached a screenshot of one of the production clusters showing a reasonable query load, with Broker and Subquery (Historical + Peon) query times ... If the latter -- on the AWS "Imply Hybrid' environments, with a beefy enough client I have measured query latency difference between what the Broker reports as query time and what the client sees, at best in the 8-9ms range. This is with a high QPS test.

I meant query latency when users/customers query data from druid since it is exposed via API where customers can select timeframe