3 ms·
What I don't understand is since it's the same service can't we just run a performance test, determine the ballpark of the needed resources and just have them c
by ahmadnoid 3y ago
What I don't understand is since it's the same service can't we just run a performance test, determine the ballpark of the needed resources and just have them capped like in k8s? why would some servers be more powerful than others in the first place?
- cdot2 3y agoSome requests may require utilizing different services than others.
- samwho 3y agoYeah, this for sure. Traffic mix changes over the life of an application. In k8s performance isolation certainly isn't a solved problem, and noisy neighbours can affect the "power" of a server. Any change to the application at all could upset the delicate balance of its performance. You would need to be doing this sort of profiling on every change, using up to date and accurate production traffic. It ends up being much easier to put some smarts in your load balancer.
- mutagen 3y ago‘Noisy neighbor’ VMs and different underlying hardware specs on the ‘same’ instance configuration are two examples that come to mind.
- adql 3y agoOne perfectly homogenous server gets 20MB POST with image that then needs to be reencoded/cropped Other gets tiny GET for CSS. Boom, now your equal servers are inequally loaded. The beauty of leastconn is that if 1st server is loaded it will just naturally get less connections than the lucky ones that only got easy jobs