Noticeable by whom? I think it is more about throughput (noticeable in the bank account) than latency (noticeable by users). If you have a fleet of servers, you need 10-25% less servers.
I think that’s the best case scenario. In all likelihood, your service is bottlenecking on I/O, and you probably still need a bunch of idle servers (to handle spikes in traffic) unless your traffic pattern is very steady or flexible (e.g., batch workloads). So even considering throughput, a 10-25% bump isn’t yielding anywhere near 10-25% reduction in servers/costs except perhaps in exceedingly rare cases.
If your service isn’t bottlenecks on I/O, then you probably care a lot more about latency than throughout anyway, or so I would think.
That’s barely noticeable, isn’t it?
Noticeable by whom? I think it is more about throughput (noticeable in the bank account) than latency (noticeable by users). If you have a fleet of servers, you need 10-25% less servers.
I think that’s the best case scenario. In all likelihood, your service is bottlenecking on I/O, and you probably still need a bunch of idle servers (to handle spikes in traffic) unless your traffic pattern is very steady or flexible (e.g., batch workloads). So even considering throughput, a 10-25% bump isn’t yielding anywhere near 10-25% reduction in servers/costs except perhaps in exceedingly rare cases.
If your service isn’t bottlenecks on I/O, then you probably care a lot more about latency than throughout anyway, or so I would think.