Messages in this thread | | | Date | Wed, 5 May 1999 16:22:52 -0400 (EDT) | From | Steve Willer <> | Subject | SMP spinlock contention -- anybody have numbers? |
| |
I've had a difficult time trying to find the kernel-level overhead of various SMP configurations. In my case, I'm particularly interested in database-style loads (heavy network, few connections/s, large files, large RAM) and how scalable Linux on Intel SMP might be. I'm particularly interested in spinlock contention, as that nicely separates kernel-level overhead from hardware- and user-level overhead.
Andrea Arcangeli's latest patches include a spinlock-measuring option, if anyone wants to measure this now. It's at ftp://e-mind.com/pub/andrea/kernel/ and includes a bunch of other stuff, so use with care (Andrea, can you please separate out the spinlock profiling patch?).
I know some people think it's only good up to 2, and others think it's fine up to 4, and few think it's any good beyond that, but does anybody have any measured comparisons, with actual numbers? I don't have any quad-CPU machines around to test. I realize the overhead depends on the application, but I don't even have a general idea of what to expect.
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |