There’s some scaling constraints using stdout with kubernetes. That logging load hits the kublet/control plane. This is an issue for those already pushing k8s up to the edge of its current scalability.
Buying 1 Mibps means you can use up to 1Mibps. Paying $/GiB means you can use 100s of Gibps for a split second and only pay the $/GiB fee.
In the past i did a TCO comparison for a large spikey workload where the $/GiB cost even at list was cheaper than allocating enough bandwidth for peak events.
At small scale the $/Mibps likely comes out cheaper, but for any large company sitting on 100s of Gibps of excess capacity for an event that occurs once or twice a year, or a few hours per day is very costly and might eat up savings from p99/p95th billing.
Many bay area tech companies decided to push Diversity representation as a top issue. Many achieved results by lowering the bar for diverse candidates to get entry level positions.
3 years later, those same diversity candidates start to notice that on AVERAGE their diversity-segment gets paid less on average that white/asian males.
BUT what the statistics show is that the bar for white/asian males happened to be higher, because the diversity initiatives prevented them from being accepted for lower-tier jobs...the same jobs the company skewed towards diverse candidates. As such, they also were paid more on average, not due to discrimination but because their group was discriminated in getting the lower-paying jobs in the company.