Nice MPD?

[nbpf]

Does it make sense, on a dedicated music server running MPD as a main process, to start MPD with a negative nice value? If so, what is a meaningful range of values? Thanks, nbpf

[max]

Priorities are relative to other processes, and if there's no other process, the nice value has no effect.

[nbpf]

Priorities are relative to other processes, and if there's no other process, the nice value has no effect.

Well, of course there are a number of system processes running in the background ... A somehow similar question is whether it would make sense to run MPD on a specific core to avoid thread swapping.

[max]

That would mean that all of MPD's threads share one core and will not benefit of multiple cores. Why would you want to do that?

[nbpf]

That would mean that all of MPD's threads share one core and will not benefit of multiple cores. Why would you want to do that?

It has been suggested in https://sites.google.com/site/computera ... ng-up-alsa. I thought I would give it a try. Thanks, nbpf

[max]

I believe it's a bad suggestion. You need a good reason to force a multi-thread process like MPD to use only one CPU core total, and the web site doesn't explain any good reason.

But, oh, that guy is talking about jitter. In digital audio processing. Oh, you fell to audiophile voodoo again!

[nbpf]

I believe it's a bad suggestion. You need a good reason to force a multi-thread process like MPD to use only one CPU core total, and the web site doesn't explain any good reason.

Thanks Max! The alleged reason is a perceived improvement in sound quality. Whether this is plausible or not, I do not know. Best. nbpf

[nbpf]

I believe it's a bad suggestion. You need a good reason to force a multi-thread process like MPD to use only one CPU core total, and the web site doesn't explain any good reason.

But, oh, that guy is talking about jitter. In digital audio processing. Oh, you fell to audiophile voodoo again!

I haven't actually read anything about jitter in the reported link and I have honestly no idea what kind of audiophile voodoo I might (again?) have fallen into! Someone has reported an observation and I thought I would repeat the experiment. What's wrong with that? I have no obvious reasons to believe the outcome will confirm or confute those observations. I do not know what you exactly mean by "good resons" but in massive parallel computations -- for instance in numerical weather prediction or industrial applications of finite elment methods -- it is not unusual to require a scheduler to run a process on a given core. Sometimes, just because experience has shown this to be more effective. Modern multi core architectures are complicated beasts. It often makes perfectly sense to take into account empirical evidences even though one does not have a fundamental explanation for such evidences. A typical example is superlinear speedup of parallel computations. This can be demonstarted in practice and explained theoretically in terms of cache effects. But it would be difficult to predict the optimal number of threads for a specific problem size on the basis of "good reasons". Here it simply makes sense to rely on empirical evidences. Best, nbpf