Messages in this thread |  | | Date | Fri, 19 Oct 2001 09:50:10 -0700 (PDT) | From | Davide Libenzi <> | Subject | Re: [PATCH] A try for a more fair scheduler ... |
| |
On Fri, 19 Oct 2001, Hubertus Franke wrote:
> Seems like a good approach. Could that be made more flexible. > Architectures today expose cache miss numbers, which > through simple markovian chains approximation allow a much better estimation > of the cache footprint then inferring from time. > Any chance to incoporate something like > this into your cost function flexibly or is this just too <way out there> ?
I just wanted to code a solution that is compatible with all architectures and that has the lower cost in terms of i-d/cache. That's why i used jiffies instead of get_cycles() and time instead of perf counters. The bottom line is that having to choose between a process that is run for 10 ms and one that is run for 150 ms, by choosing the 150 one we have a very good probability to pick up the one that has a greater footprint. Coding more complex solutions will have the effect to add an extra cost that will _probably_ vanish the better behavior given by the patch.
- Davide
- 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/
|  |