Kai Vehmanen wrote: > > This is great news. Linus has now added Robert Love's kernel preemption > patch to the 2.5 tree! > > http://www.linuxdevices.com/news/NS3989618385.html > > --cut-- > "The addition of kernel preemption to the 2.5 tree is a substantial > feature which will provide benefit to a range of uses," Love told > LinuxDevices.com.
Alas, Robert protesteth too much. As I'm sure those on this list have noticed (I've only rejoined recently), the preemptive kernel is basically useless at this time. Worst-case latencies are still in the 50-100 millisecond range, and the average is only moderately improved. The comments and reactions in the popular press are pretty amazing to someone who has actually performed the measurements, trust me. I recommended to Linus that the preemptive kernel patch be merged, because it provides a sane basis upon which to start working on the *hard* parts of providing good latencies from the kernel. That work hasn't started yet. Probably a lot of it will come from myself. So you can expect kernel 2.6 to be useful for the applications which people on this list are interested in. But 2.6 is some time away. Maybe years. At this time the only solutions are a patched kernel with the low-latency patch, or the preempt patch and the "lock-break" patch, which is a low-latency derivative. There is hope - at present, RedHat's rawhide kernel, at ftp://ftp.redhat.com/pub/redhat/linux/rawhide/i386/RedHat/RPMS/ contains the full-on low-latency patch. Whether that is permanent I do not know. -
