Messages in this thread | | | Date | Thu, 3 Aug 2000 17:24:25 +0100 (BST) | From | Tigran Aivazian <> | Subject | Re: [benchmark] 2.4.0-test6-pre1 PAE vs non-PAE |
| |
Hi Ingo!
On Thu, 3 Aug 2000, Ingo Molnar wrote: > the 3% performance drop
that is 6, not 3. 129.6/136.5 = 0.94945054945054945054 ~ 0.94 (ok, 0.95 but no way it is 0.97 :)
> is mainly due to two values, 'Pipe-based Context Switching', >'Process Creation', and 'Execl Throughput'.
that is 3, not 2.
the rest of your mail makes perfect sense, thank you.
> The fork() and > exec() result is understandably worse with PAE, because the 'density' of > page-tables is half of that of non-PAE page-tables (ie. twice as much has > to be copied), plus there are 3 user-space root page tables instead of the > 1. (which have to be zeroed, so this shows up big time.) > > Another performance problem is likely the amount of LOCK-ed instructions > done within the PAE include-files - some of that is unnecessery as David > S. Miller noticed. > > otherwise the PAE kernels show no performance drop in 'typical' user-space > stuff. But yes, you dont want to use it on a box with less than 4GB RAM.
Regards, Tigran
- 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/
| |