Messages in this thread | | | Date | Sat, 15 Nov 2003 20:14:18 +1100 | From | Nick Piggin <> | Subject | Nick's scheduler v19 |
| |
http://www.kerneltrap.org/~npiggin/v19/
Quite a few changes. Although inter node and intra node migration is generally reduced in my patches, it seems probably context switching is hurting more. This would explain the decreased system time and increased user time - the user code is thrashing the cache more.
I haven't addressed that problem yet, although I have removed the last (I think) major source of cacheline dirtying between CPUs. I say major, but it was difficult to measure a difference on a 16-way NUMAQ.
For desktop users, microsecond accounting. This currently incurs a bit of unneeded overhead because I have to convert from 64-bit nanoseconds to 32-bit microseconds everywhere. I'm using microseconds because a 32-bit value can store a maximum interval of about 4 seconds in nanoseconds, which causes me all sorts of overflow problems.
I'm pleased to have had good results with HZ timing up until now. I think this helps interactivity but I haven't had many problems with it myself. Its against mm3 but I can do a patch against Linus' tree.
Best regards, Nick
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |