Messages in this thread |  | | Date | Wed, 23 Aug 2000 19:38:14 +0200 | From | "Andi Kleen" <> | Subject | Re: SCO: "thread creation is about a thousand times faster than on native Linux" |
| |
On Wed, Aug 23, 2000 at 06:29:28PM +0100, Tigran Aivazian wrote: > It is plausible that individual _lwp_create(2) is faster than individual > clone(CLONE_VM) one (possibly between 5x and 10x times, the 1000x figure > is ridiculous). However, one should look a bit further and ask if native > kernel-level UW7 lwps are delivering the same set of rich functionality as > Linux clone'd tasks. IMHO, they simply don't. So one has to add all the > overhead of turning a "raw" lwp into a useful and useable thread, e.g. as > used by userspace thread concept. If you do that, no doubt you will find > that Linux performance is greater than of any commercial alternatives.
The main problem Linux clone has over LWPs is the extensive memory resource usage (~8.5K on x86, 16+somethingK on 64bit for the kernel stacks)
To solve that it would be very nice to have Mach like continuations in Linux (multiple kernel threads sharing a kernel stack)
Also the LinuxThreads library does horrible things at pthread_create time, like doing a full context switch to the manager thread and cloning from there (mostly to work around signal/waitpid() problems in the kernel)
-Andi - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org Please read the FAQ at http://www.tux.org/lkml/
|  |