So, the real problem isn’t noted until towards the end which is that these methods require keeping some state, thus making the requests more expensive. It’s a

So, the real problem isn’t noted until towards the end which is that these methods require keeping some state, thus making the requests more expensive. It’s a good trade off when throughput is low and the impact of lumpy traffic is high. At higher volumes, random allows stateless load balancing. Use the tool that makes sense for your project. Even at the same system I’ll have some services that operate at low volumes with high latency and others with high volume and low latency.