In the project I've been working on I recently came across a situation where the service was hammered by updates and it started struggling under the load of simultaneous client (predominantly read) requests. The solution was maintain separate thread pools for updates and client requests. The service needs to be available, but we can accept delays in updates. This inspired me to play around with thread pools - which for some reason I've never done before - in a small demonstration.
If you run CommonThreadPoolExperiment you'll see that the number of completed read processes converges to that of the write processes. On my machine, around 10 seconds.
With SeparateThreadPoolsExperiment the numbers are starkly different, the reads are unperturbed by the much slower writes.
