> Not really, during load your reserved cpu will now have to wait longer
> for shared resources instead of helping make progress, bringing down the
> performance of all your applications including the 'realtime' one.
Of course you are right in the general case. But as long as one has correctly
sized the load so that it does NOT exceed the dedicated CPUs, and one has
reserved the right resources, then it can help the stability of the soft-
realtime applications that are of interest to me.
So, in my case, using dedicated raw-ish disks, pinned memory, dedicated CPUs
and the understanding that the kernel has absolute priority over userspace (and
so is never a worse bottleneck than usual), it all works. Also, my application
maximises the probability of success by explictly managing the resources which
*are* shared between the time-sensitive code and the relevant "joe-random" code.
Can I be certain that there is no shared lock or anything else in the whole
kernel? No, but I'm prepared to make that probabalistic tradeoff (backed via
extensive testing) rather than have to go to hard-realtime.
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to email@example.com
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/