On Thu, Oct 18, 2018 at 12:33:32PM +0200, luca abeni wrote: > Hi Peter, > > On Thu, 18 Oct 2018 11:48:50 +0200 > Peter Zijlstra <pet...@infradead.org> wrote: > [...] > > > So, I tend to think that we might want to play safe and put some > > > higher minimum value for dl_runtime (it's currently at 1ULL << > > > DL_SCALE). Guess the problem is to pick a reasonable value, though. > > > Maybe link it someway to HZ? Then we might add a sysctl (or > > > similar) thing with which knowledgeable users can do whatever they > > > think their platform/config can support? > > > > Yes, a HZ related limit sounds like something we'd want. But if we're > > going to do a minimum sysctl, we should also consider adding a > > maximum, if you set a massive period/deadline, you can, even with a > > relatively low u, incur significant delays. > > I agree with this. > > > > And do we want to put the limit on runtime or on period ? > > I think we should have a minimum allowed runtime, a maximum allowed > runtime, a minimum allowed period and a (per-user? per-control > group?) maximum allowed utilization.
I was talking about a global !root max-u, but yes the cgroup max-u makes definite sense as well. > I suspect having a maximum period is useless, if we already enforce a > maximum runtime. Probably; yes. The asymmetry is unfortunate of course. > > That is, something like: > > > > TICK_NSEC/2 < period < 10*TICK_NSEC > > As written above I would not enforce a maximum period. I'm confused: 'period < 10*TICK_NSEC' reads like a max to me. (irrespective of the argument on wether the max should be HZ related; and I think you and Juri made good argument for it not to be) > > and/or > > > > TICK_NSEC/2 < runtime < 10*TICK_NSEC > > I think (but I might be wrong) that "TICK_NSEC/2" is too large... I > would divide the tick for a larger number (how many time do we want to > allow the loop to run?) It depends on how strict we want to enforce the no-interference rule. The smaller we make this, the less accurate we enforce, the worse the interference between tasks. Note that we're only talking about a default; and HZ=100 is daft in any case. > And I think the maximum runtime should not be TICK-dependent... It is > the maximum amount of time for which we allow the dealdine task to > starve non-deadline tasks, so it should be an absolute time, not > something HZ-dependent... No? Agreed. > > Hmm, for HZ=1000 that ends up with a max period of 10ms, that's far > > too low, 24Hz needs ~41ms. We can of course also limit the runtime by > > capping u for users (as we should anyway). > > Regarding capping u for users: some time ago, with Juri we discussed > the idea of having per-cgroup limits on the deadline utilization... I > think this is a good idea (and if the userspace creates a cgroup per > user, this results in per-user capping - but it is more flexible in > general) Agreed, that makes sense.