3 years ago

Category: Data, Infrastructure

It covers some of the things an engineer building out and supporting an application needs to be mindful of, including CPU and memory characteristics to look out for when an application is running, and guidance on how to design and implement the synthetic metric. What do you do with a server that costs a lot of money and does not do much work? It sounds like the start of a (bad) joke, the answer, obviously, is to understand why its not doing much, and if there’s no good reason, save your cost centre some money and get rid of it.

It’s impossible to tell on its own, the server may, for example, be hosting a web-queue-worker application that reads a relatively small number of requests over the network but then spends are large amount of time crunching through each request, driving up CPU usage, and then returns a relatively small response — in this case, network traffic is low, but not a bad thing.

For example, it may be known in advance that a large number of apps are running with a fixed memory footprint, and CPU usage is going to be used as the dominant measure of how active (busy) the applications are.